Evidence-weighted experimental prioritization of unknown bacterial genes

Functional Dark Matter — Experimentally Prioritized Novel Genetic Systems describes an evidence-weighted framework for converting unknown bacterial genes into experimentally testable hypotheses by combining fitness, essentiality, conservation, pathway, environmental, genomic-context, and annotation evidence. Truly Dark Genes — What Remains Unknown After Modern Annotation? refines this framework by separating annotation-lag genes from genes that remain hypothetical after modern reannotation and by adding a clue-based prioritization of the residual unknown set. Conservation vs Fitness — Linking FB Genes to Pangenome Clusters further refines the conservation axis by directly linking Fitness Browser genes to KBase pangenome clusters and testing essentiality against core, auxiliary, and unmapped conservation categories. discoveries supports the framework’s evidence-triangulation premise: across 47 testable dark-gene clusters, 29 (61.7%) showed directional concordance between laboratory fitness conditions and carrier-genome environments, with Fisher’s combined p=0.031, while independent NMDC validation confirmed all 4 pre-registered abiotic predictions and all 7 pre-registered trait predictions. ₁ The corpus also cites earlier genome-wide mutant-phenotype work, “Mutant phenotypes for thousands of bacterial genes of unknown function” (Nature 557(7706):503; PMID 29769718), as a reference point for this kind of phenotype-first annotation. ₂

Recommended follow-up from the dark-gene reports

The functional dark matter report’s future-directions list recommends a dual-route campaign. Route A (NB09, evidence-weighted) identifies 42 organisms covering 95% of composite priority and would start with MR-1 stress/nitrogen screens and E. coli crispri for targeted hypothesis testing. Route B (NB11, conservation-weighted) identifies 42 organisms covering 95.6% of importance-weighted priority and would start with S. meliloti, P. putida and MR-1 broad phenotypic screens for novel function discovery. The two routes share 39 organisms. The report calls them complementary: Route A yields mechanistic insight for genes that already carry hypotheses, and Route B targets broadly conserved true knowledge gaps. The 95% and 95.6% figures use different denominators from each other and from the extended covering set’s orthologous-group (OG) coverage described below. All of these are proposed experiments, not completed screens. ₃

The report also lists targeted validation of the top 5 fitness-active candidates, 4 in MR-1 and 1 in P. putida N2C3, which it calls immediately testable via RB-TnSeq (random-barcode transposon sequencing) under their predicted condition classes. AO356_11255 (D-alanyl-D-alanine carboxypeptidase prediction, EamA domain) would be tested under nitrogen limitation. MR-1 202463 (YGGT domain, T5 Dawn, composite score 0.698) would be tested under multiple stress conditions. The K03306 paralog family MR-1 199738/203545/202450 would be tested under nitrogen limitation with single/double mutant comparisons. This supports the screen designs described later on this page, but the functions remain predictions until the screens are run. ₃

For essential dark genes, the top 50 essential candidates (NB07) come with specific CRISPRi (CRISPR-interference knockdown) experiment designs. The highest-priority targets are Keio:14796 (YbeY domain, score 0.875, predicted ion transport), MR-1:200382 (RimP_N/DUF150_C, score 0.874, predicted ribosome assembly) and Koxy:BWI76_RS08540 (OmpA/TIGR02802, score 0.865, predicted cell division). The covering set assigns 8,900 essential genes to specific organisms for CRISPRi-based approaches. The listed roles are predicted functions, not knockdown results. ₃

The report proposes protein structure prediction for the top 100 candidates and especially the 1,853 T5 Dawn genes (nearly characterized). It says AlphaFold2 structure predictions could confirm functional hypotheses and distinguish between DUF families. It describes the 4,273 T1 Void genes as lacking all evidence and calls structure-based function prediction the only remaining computational inference strategy for them. Any such assignment would therefore remain computational inference rather than experimental validation. ₃

A future-directions passage says the current covering sets draw only from 48 FB organisms, which are 77% Pseudomonadota. It proposes an extended tractable list of 25 additional organisms with published TnSeq/CRISPRi resources from Bacillota (6), Actinomycetota (2), Campylobacterota (2) and additional Pseudomonadota (15), so that kingdom-level OGs (55.9% of dark gene OGs) could be studied in Gram-positive, Actinobacterial and Campylobacterial contexts. This refines the extended-list description later on this page, which names only the three non-Pseudomonadota phyla. The passage also says the extension requires a pangenome Spark query, an extension of the NB11b pipeline, yet the same report elsewhere describes a completed extended covering set (NB11c). The report does not reconcile the two statements, so both are preserved as reported. ₃

The truly dark analysis uses its own multi-criteria score: fitness importance, annotation clues, ortholog breadth, genomic context and tractability, with a maximum score of 12. It ranks all 6,427 truly dark genes and identifies 100 top candidates (scores 8–10) across 19 organisms. Methanococcus_S2 (29), DvH (13), Methanococcus_JJ (13) and MR-1 (8) contribute the most. Of the 100, 34 are essential, 53 are in operons and 30 are in ICA (independent component analysis) fitness modules, which group genes with coordinated fitness profiles. ₄ This scale is not comparable with the functional dark matter ranking, whose six dimensions are each scaled 0–1. ₃

• PV4/5210953 (score 10): motility phenotype (|f| = 5.5), in an operon with TatC (Sec-independent translocase), ICA module M016.
• ANA3/7026383 (score 9): nitrogen-source phenotype (|f| = 8.6), in an operon with an ABC transporter, ICA module M018.
• DvH/206658 (score 9): stress phenotype (|f| = 5.4). eggNOG suggests trehalose synthase despite the hypothetical annotation, which the report flags as a potential mis-annotation or novel variant that still needs resolution.
• Methanococcus_S2/MMP_RS06570 (score 9): 8 annotation clues, including DUF190 and COG-T (signal transduction in the Clusters of Orthologous Groups, COG, classification), in an operon with fluoride efflux transporter CrcB. ₄

The truly dark clue tiers refine the clue-matrix summary below. Tier 1 (no clues) holds 246 genes (3.8%). Tier 2 (sequence identifiers only, UniRef50/RefSeq) holds 3,867 (60.2%). Tier 3 (partial function from Pfam, COG or eggNOG) holds 711 (11.1%). Tier 4 (fitness, essential or module signal but no functional annotation) holds 1,603 (24.9%). The report calls Tier 3 and Tier 4 genes (2,314 total) the most promising for functional characterization because their partial clues can narrow experimental hypotheses. ₄

The truly dark candidate list is incomplete by the report’s own gap assessment. Of the dark genes, 17,479 (31%) lack pangenome links and could not be assessed by Bakta. Extrapolating the 16.3% truly-dark rate, the report estimates ~2,841 additional truly dark genes among them, including 2,208 with strong fitness phenotypes. It states that the ranked list covers ~69% of the estimated total truly dark gene population and calls this a 31% coverage gap. These are extrapolated estimates, not counts of assessed genes. ₄

The truly dark report’s proposed next steps are not completed evidence:

• run AlphaFold2/ESMFold on the top-100 candidates and search Foldseek for remote structural homologs, which it calls the most promising path for sequences with no detectable sequence homology;
• run growth assays in predicted conditions, such as PV4/5210953 in motility assays and ANA3/7026383 under nitrogen limitation, with Mobile-CRISPRi (Peters et al. 2019) suggested for rapid testing across organisms;
• run a Methanococcus deep-dive, motivated by the report’s statement that archaea contribute 55% of truly dark genes, using high cofitness signals (r > 0.97) to map operonic structure and predict functions for archaeal-specific hypothetical proteins. ₄

Candidate pools from adjacent projects

The metal fitness atlas supplies a phenotype-first candidate pool from a single evidence stream. It identifies 149 novel metal biology candidates: gene families with conserved metal fitness phenotypes across ≥2 organisms but lacking full functional annotation. They comprise 89 truly unknown (no annotation), 43 with DUF/UPF domains (known structural domain, unknown metal role) and 17 with partial functional hints such as transporter or hydrolase. The report states that these predictions rest solely on cross-species fitness data, and it proposes PaperBLAST, InterPro and AlphaFold structural prediction to characterize the 89 truly unknown and 43 DUF-domain families. Under the evidence-weighting logic of this page, they are single-axis candidates whose functions remain to be characterized. ₅

Condition coverage also shapes this pool. The report’s text says two organisms have disproportionately many novel metal candidate genes (54 each) but have been tested against only 1 metal (iron). Its table names them as methanococcus-maripaludis S2 and JJ and lists the 54 as novel OGs for each strain. The report does not say whether genes or OGs is the intended unit, so both formulations are preserved and the distinction remains unresolved. For S2 the report proposes Ni, Co, Cu, Zn, W and Mo, noting that archaea are underrepresented in metal tolerance studies. For JJ it proposes the same panel for cross-strain comparison with S2. The report’s expected-impact statement covers its full proposed campaign, which also includes rare-earth-element (REE) experiments and experiments in additional organisms, not the Methanococcus panel alone. If those experiments were performed, it projects that the atlas would expand from 14 to 16+ metals (adding REEs), from 24 to 27+ organisms with fitness data, and from 1-2 organisms for critical metals to 5+, and that the 149 candidates could be directly validated in organisms where they are predicted to be important. These are conditional projections, not observed results, and they support fitness-condition-coverage-prioritization-bias. ₅ The same strains recur elsewhere: Methanococcus_S2 contributes 29 and Methanococcus_JJ 13 of the truly dark top 100 ₄, and Route B uniquely selects Methanococcus_JJ and Methanococcus_S2. ₃ This makes the Methanococcus metal panel a newly testable point of convergence across three prioritization schemes, although no report links the specific genes. _5,4,3

The metal specificity analysis ranks the most metal-specific candidate families. ucp030820 (OG01015, 3 orgs, 7 metals) had 2/3 metal-specific records (67%), with mean sick rate 0.021. yebc (OG01383, 11 orgs, 6 metals) had 7/12 (58%), with mean sick rate 0.056. duf1043 yhcb (OG03264, 6 orgs, 5 metals) had 3/6 (50%), with mean sick rate 0.054. Across all 149 novel metal candidate families, 45.6% have a dominant specificity of metal-specific, compared with 58.2% for annotated families (Fisher exact OR=0.60, p=0.003). The comparison is qualified by the novel set’s concentration in deeply profiled organisms. This refines the metal pool: unknown metal-phenotype families are less often metal-specific than annotated ones. ₆ The identifier OG01383 is also assigned to a pH-associated ortholog group in the functional dark matter roadmap. ₃ Neither report states whether the two projects share ortholog-group definitions, so the coincidence should not be read as a link. _6,3

For YebC, the report cites Ignatov et al. (2025), Nature Communications, PMID 40624002. That study showed YebC to be a ribosome-associated translation factor for proline-rich proteins that resolves ribosome stalling at polyproline motifs. This is cited literature, not a result established in the analysis. The report’s proposal that YebC is specifically required under metal stress, because demand for proline-rich metal transporters creates a translation bottleneck, is a hypothesis it did not test. It also proposes using AlphaFold predictions to identify metal-binding sites in the top metal-specific candidates (UCP030820, YebC, DUF1043), which is proposed work, not a reported result. ₆

Literature linkage offers a separate prioritization input. The cross-database analysis found 129,823 VIMSS cross-references connecting PaperBLAST to the Fitness Browser. This creates a path from literature coverage to experimental fitness phenotypes, which the report describes as a resource for identifying functionally important but understudied genes and which bears on research-attention-inequality. ₇

The plant microbiome analysis contributes a mechanistically dark rather than annotation-dark pool. It presents 50 novel OGs enriched in plant-associated species after phylogenetic control as candidates for previously unrecognized plant-interaction functions. By COG category, 19/50 (38%) are unknown function S, 7/50 (14%) energy production C, 6/50 (12%) inorganic ion transport P, 4/50 (8%) carbohydrate metabolism G and 3/50 (6%) secondary metabolite biosynthesis Q. The report’s reading that this indicates rhizosphere nutrient competition rather than plant-specific signaling is suggestive, not experimentally established. All 50 OGs have annotated function from that project’s NB09 but lack mechanism-level proposals. The report names COG3569 (cog3569), COG1845, COG0843 and COG0316 for AlphaFold structural prediction or pathway-database cross-referencing. It also identifies a 17-genera core rhizosphere set and Tier-1 markers as candidate organisms and gene targets: nitrogen fixation, ACC (1-aminocyclopropane-1-carboxylate) deaminase and T3SS (type iii secretion system). This extends the framework: an annotation label does not remove the need for mechanism-level experiments. ₈

Scope and evidence base

Across 48 organisms and 228,709 genes, the analysis identified 57,011 dark genes—hypothetical, DUF, or otherwise uncharacterized genes—representing 24.9% of the catalog. ₃ The proportion of dark genes ranged from more than 35% to less than 15% among organisms, which the report interprets as annotation-depth variation rather than established variation in functional content. ₃

Of the 57,011 dark genes (24.9%) across 48 organisms and 228,709 genes, fitness and essentiality supplied the main experimental evidence streams: 7,787 dark genes had strong fitness effects, defined as absolute fitness of at least 2 in at least one condition, and 9,557 were classified as essential because no viable transposon mutants were recovered. ₃ Together, 17,344 dark genes had experimentally measurable phenotypes under the available assays. ₃ This supports condition-specific-fitness and gene-essentiality by showing that unknown sequence annotation does not imply an absence of experimentally accessible phenotype. Discoveries further supports this interpretation at broader scale: 25,271 genes (17.8%) were true trade-off genes, important (fit < -1) in some conditions and burdensome (fit > 1) in others, showing why condition-specific measurements should remain separate evidence axes rather than being collapsed into a single essentiality label. ₁

The evidence base was uneven across gene classes: 39,532 of the 57,011 dark genes (69.3%) had pangenome links, 12,686 were accessory, 511 were both accessory and strongly fitness-active, and 6,142 belonged to independent-component-analysis (ICA) fitness modules, which group genes with coordinated fitness profiles. ₃ The census gives its own percentages. It identifies 57,011 dark genes (24.9% of 228,709 total), of which 39,532 (69.3%) link to the pangenome, 6,142 (10.8%) belong to ICA fitness modules with function predictions, 7,787 (13.7%) show strong fitness effects (|fit| ≥ 2) and 9,557 (16.8%) are essential. The census also reports that the intersection most relevant to biogeographic analysis, accessory dark genes with strong fitness, contains 511 genes. The report concludes that one quarter of bacterial genes remain dark but that the problem is tractable, because 17,344 have measurable fitness or essentiality phenotypes and 39,532 connect to the pangenome. ₃ Stress conditions, including metals, oxidative stress, and osmotic stress, produced the largest concentration of strong dark-gene phenotypes, followed by carbon-source and nitrogen-source conditions. ₃ The residual truly dark subset is more selective: among genes with strong phenotypes, stress conditions were reported less often for truly dark genes than for annotation-lag genes, while mixed-community, iron, nutrient time-series, and rich-media signals were reported as enriched. ₄

The new pangenome linkage supports using conservation as an evidence axis but also calibrates its interpretation: 44 of 48 Fitness Browser organisms mapped to pangenome species clades, producing 177,863 gene-to-cluster links with 100.0% median protein identity and 94.2% median gene coverage; 33 organisms were retained for downstream analysis after coverage and identifier exclusions. ₉ Among 148,826 protein-coding genes in those organisms, 27,693 were putative essential genes, or 18.6% overall, with per-organism values ranging from 12.9-28.9%. ₉ Essentiality was inferred from RB-TnSeq, a random-barcode transposon sequencing approach, under the library construction and growth conditions represented in the Fitness Browser. ₉

Discoveries refines the evidence base by showing that unknown-gene status is often a property of annotation coverage rather than absence of homologous information. Across 27,690 species, 55.9% of dark-gene ortholog groups were kingdom-level, with species counts ranging from 1 to 27,482 and median 135; in a 57,011-gene darkness spectrum, only 4,273 genes (7.5%) were T1 Void, with zero evidence, whereas 22,500 (39.5%) were T4 Penumbra with 3–4 converging evidence lines (fitness phenotype, domain annotation, ortholog group, module membership). The report concludes that the key bottleneck is targeted experimental validation rather than computational inference. ₁

Evidence-weighted prioritization

The six-axis prioritization score combined fitness importance, cross-organism conservation, inference quality, pangenome distribution, biogeographic signal, and experimental tractability. ₃ It ranked 17,344 phenotype-bearing dark genes, and the top 100 candidates, with scores ranging from 0.624 to 0.715, spanned 22 organisms. ₃ Shewanella MR-1 contributed 25 of the top 100 candidates, Pseudomonas putida N2C3 contributed 18, and Marinobacter contributed 9. ₃ Of these candidates, 82/100 had high-confidence functional hypotheses supported by at least 3 evidence types, and 85/100 had module-based predictions. ₃ In addition, 97/100 had domain annotations. The report presents the output as 100 ranked candidates with specific functional hypotheses and suggested experiments. ₃ Each of the six dimensions is scaled 0–1. The composite is meant to translate diverse evidence types into a single prioritization that an experimentalist can act on, so the top-ranked genes are those where the most evidence converges and the experimental path is clearest. ₃ The report describes the 17,344 ranked genes as phenotype-bearing and says the top candidates show converging fitness, conservation, co-regulation and domain evidence. Its census defines the 17,344 as the 7,787 genes with strong fitness effects together with the 9,557 essential genes, not as two separate populations. ₃

The score is therefore best interpreted as a decision aid rather than a direct probability that a gene has a particular function. ₃ Rank stability was high for overall configurations, with rank correlations remaining ρ > 0.93 across six alternative configurations, but only 64% of the original fitness-active top 50 remained under conservation-dominant or drop-tractability settings. ₃ Essential-gene rankings were more sensitive: essential-gene top-50 retention was 36% when tractability was dropped and 48% when neighbor context was dropped. ₃ The report notes that both frameworks use arbitrary expert-assigned weights. In counts, the fitness-active conservation-dominant and drop-tractability configurations each retained 32/50 original top candidates. The essential-gene configurations retained 18/50 without tractability and 24/50 without neighbor context. ₃ Robust rank indicators identify 18 fitness-active genes that stay in the top 50, and 35 that stay in the top 100, across all 6 weight configurations. The corresponding essential-gene counts are 6 always-top-50 and 19 always-top-100 genes. The report calls these always-top candidates the most defensible targets for experimental follow-up. ₃ These sensitivity results support experimental-prioritization-of-functional-dark-matter as a framework for transparent trade-offs rather than a fixed universal ranking. ₃

How breadth is measured also affects the conservation axis. The report describes dark-gene clusters as ranging from clade-restricted (single phylum) to widespread (3+ phyla), and it treats widespread unannotated families as the highest-priority unknowns. However, it also notes that this eggNOG-derived breadth classification is coarse-grained: 99.9% of dark gene clusters (30,721 of 30,756) map to “universal” breadth, so the classification does not discriminate among candidates. ₃ Replacing the binary breadth classification with a continuous species-count metric (min(n_species / 20, 1.0) × 0.5) produced rankings highly correlated with the original (Spearman ρ = 0.982). Even so, top-50 overlap was only 62% and top-100 overlap 58%. This refines the high overall rank stability reported above: candidate-level lists change materially even when global correlations remain high. ₃

The conservation-versus-fitness analysis supports the use of essentiality as a prioritization signal while showing that it is not equivalent to conservation: essential genes were 86.1% core versus 81.2% for non-essential genes, with a median odds ratio of 1.56; 18 of 33 organisms showed significant enrichment after Benjamini-Hochberg false discovery rate (BH-FDR) correction at q < 0.05. ₉ Thus, viability-required genes are generally more conserved within species, but the enrichment is modest and most genes in well-characterized bacteria are core regardless of essentiality. ₉ Discoveries supports this caution across 48 bacteria: 859 of 17,222 ortholog families (5.0%) were universally essential, 4,799 (27.9%) variably essential, and 11,564 (67.1%) never essential; universally essential families were 91.7% core versus 80.7% for non-essential genes, while orphan essential genes were only 49.5% core. ₁

The truly dark analysis supports this evidence-weighting principle but narrows the target population. Of 57,011 dark genes, 33,105 were reclassified by Bakta and 6,427 remained hypothetical in both pipelines among 39,532 pangenome-linked genes; 17,479 lacked pangenome links and could not be assessed. ₄ The persistent set was ranked using fitness importance, annotation clues, ortholog breadth, genomic context, and experimental tractability, producing 100 candidates across 19 organisms; 34 were essential, 53 were in operons, and 30 were in ICA modules. ₄ This refines the earlier ranking by showing that an unknown-gene campaign should distinguish residual annotation resistance from database lag and should explicitly represent linkage uncertainty.

A 12-dimensional clue matrix further supports prioritizing genes with partial evidence rather than treating all dark genes as equally unknown: 246 genes (3.8%) had zero clues, whereas 2,314 genes (Tier 3 plus Tier 4) had partial functional or phenotype evidence suitable for narrowing experimental hypotheses. ₄ The result complements, rather than replaces, the earlier finding that 82/100 candidates had hypotheses supported by at least 3 evidence types. _{3 4} Discoveries supports and broadens this distinction: Bakta v1.12.0 reclassified 33,105 of 39,532 dark genes with pangenome links (83.7%), leaving 6,427 truly dark genes; only 246/6,427 (3.8%) had zero annotation clues. ₁

Discoveries refines what those clues consist of. The original dark-gene classification came from the kescience-fitnessbrowser (Fitness Browser) annotations. Only 246/6,427 (3.8%) truly dark genes have zero annotation clues. Of the 6,427, 84.7% have database cross-references and 43.5% have eggNOG hits, although 55% of COG (Clusters of Orthologous Groups) assignments are “S”/unknown. In addition, 29.3% have cross-organism orthologs and 9.5% are in ICA fitness modules. ₁ Genomic context points the same way: 41% of their neighbors are also hypothetical, and 12% lie within 2 genes of mobile elements (transposases, integrases, phage proteins). The report infers concentration in genomic islands from this pattern but does not demonstrate it. ₁

Annotation breadth itself should be treated as a composite evidence problem. Discoveries found that Bakta and eggNOG were complementary across 132.5M gene clusters: their union increased any-functional-annotation coverage to 77.3%, and Bakta rescued 11.2M clusters among 39.2M missed by eggNOG. ₁ This supports using multiple annotation pipelines before assigning experimental priority, while the fact that only 33.3% of Bakta’s 17.6M distinct UniRef50 IDs existed in the KBase Data Lakehouse UniProt identifier table refines the interpretation by making database provenance and identifier coverage part of the ranking uncertainty. ₁

From candidate genes to experiments

A greedy set-cover analysis found that 10 organism–condition experiments would address 242 of the top 500 dark genes, reported as 45.3% of the top 500. That percentage does not equal 242 of 500, and the table’s other cumulative percentages likewise do not match a 500-gene denominator. The coverage shares are therefore preserved as reported rather than recalculated. ₃ The first three proposed experiments—Shewanella MR-1 under stress, nitrogen-source, and carbon-source conditions—would address 111 candidates, reported as 20.8% (again not equal to 111 of 500). ₃ The complete ten-experiment sequence also included stress or carbon-source experiments in P. putida N2C3, Sinorhizobium meliloti, P. stutzeri RCH2, Marinobacter, P. fluorescens GW456-L13, and P. putida. ₃

The three proposed MR-1 screens would cover 111 top-500 candidates (reported as 20.8%). This is a projection rather than a completed result. It reflects MR-1’s 121 historical conditions, 587 scored dark genes and 142 genes with |fit| ≥ 4. _3,1 The proposed stress screen addresses 57 genes, led by 202463 (YGGT domain, |fit|=6.4), 203247 (GBBH-like_N, |fit|=4.6) and 203631 (|fit|=4.5). All three carry PF01145 module predictions that suggest, but do not establish, a stress-responsive membrane-protein family. The nitrogen screen addresses 31 genes, led by the K03306 paralog trio 199738 (Gcw_chp/TIGR02001, |fit|=5.5), 203545 (DUF4124, |fit|=4.0) and 202450 (Gly_transporter, |fit|=3.9); comparing single and double mutants would test functional redundancy. The carbon screen addresses 23 genes, led by 202474 (YggL_50S_bp, |fit|=7.1) and 202608 (BcrAD_BadFG, |fit|=7.3), the highest fitness magnitudes in the top 100. ₃

P. putida N2C3 is the second-highest-value organism, with 48 genes in the top 500, of which a stress screen alone would address 23. ₃ Its locus AO356_11255 is the project’s top-ranked candidate. It has a D-alanyl-D-alanine carboxypeptidase prediction, an EamA domain and |fit|=3.4 under nitrogen. It also shows the clearest lab–field signal: 80% of carrier genomes come from soil/freshwater/wastewater environments versus 8.3% of non-carriers (OR = 44, FDR = 0.093). ₃ That association is consistent with the nitrogen-utilization phenotype but does not meet the FDR < 0.05 threshold used for the within-species environmental tests. It therefore supports prioritization rather than establishing function. ₃

Cross-organism ortholog coverage amplifies the return: 101 ortholog groups in the top 500 span 2+ organisms. ₃ Examples include:

• carbon-source OG03827 (MR-1, Marinobacter, ANA3, PV4, P. fluorescens N1B4, P. simiae N2E2);
• nitrogen-source OG02907 (MR-1, E. coli Keio, K. oxytoca, P. putida, P. fluorescens N1B4, P. simiae N2E3);
• pH-associated OG01383 (MR-1, S. meliloti, C. metallidurans, P. fluorescens GW456-L13, P. putida N2C3);
• stress-associated OG03534 (MR-1, ANA3, Dinoroseobacter, PV4, Synechococcus);
• stress-associated OG01997 (MR-1, ANA3, S. meliloti, P. putida N2C3). ₃ The report proposes that stress screens in MR-1 and P. putida N2C3 would provide cross-organism concordance data for OG01997. Combined with nitrogen experiments, these screens would also test OG02907 and OG01383 across genetic backgrounds. This is a projection, not a completed test, and the source does not explain how stress or nitrogen screens would test OG01383, which its own table assigns to pH. ₃

Discoveries refines the MR-1-first ranking by showing that it depends on the weighting route. Under evidence-weighted prioritization (Route A), MR-1 contributes 25/100 top fitness-active candidates and 587 scored dark genes, and covers 20.8% of the top-500 in 3 experiments. This reflects deep condition coverage (121 conditions), a large dark gene complement and high fitness effect magnitudes (142 genes with |fit| >= 4). Under conservation-weighted prioritization (Route B), S. meliloti ranks first with 1,630 ortholog groups (OGs) and 195 kingdom-level gaps. ₁ Both covering sets select 42 organisms covering ~95% of priority and share 39 organisms. The routes are therefore complementary rather than contradictory: Route A suits targeted hypothesis testing, and Route B suits novel function discovery. ₁ Adding 25 non-Fitness Browser organisms (Bacillota, Actinomycetota, Campylobacterota) to the candidate pool selects 50 organisms spanning 6 phyla, versus 4 for Fitness Browser organisms alone. In that extended set, P. aeruginosa PAO1 ranks #1 (3,713 OGs) and M. tuberculosis reaches #6 as the first Actinomycetota. ₁ As a null result for set-cover value, the Bacillota organisms B. subtilis and S. aureus were not selected because their kingdom-level OGs were already covered by Pseudomonadota selections, although they remain valuable for studying genes in native Gram-positive context. ₁ This qualifies the taxonomic-imbalance warning below: broader phylum coverage changes which organisms are selected, but unique-OG coverage alone does not reward Gram-positive representation. ₁

This experiment-level optimization extends fitness-module-detection-sensitivity by treating condition selection and organism selection as coverage problems rather than evaluating genes independently. ₃ The approach also exposes a depth bias: Fitness Browser condition coverage is uneven, and MR-1 has 121 historical conditions, so organisms with deeper condition coverage can produce more specific phenotypes and receive higher prioritization scores. ₃

For non-essential genes, RB-TnSeq—randomly barcoded transposon sequencing—provides condition-specific fitness measurements when viable insertion mutants are available. ₃ Essential dark genes require a separate strategy because standard RB-TnSeq yields no viable knockout fitness profiles for them. ₃ The report recommends CRISPRi, or CRISPR interference for transcriptional knockdown, including Mobile-CRISPRi for less-established organisms, followed by growth measurements under standard and stress conditions. ₃ The truly dark analysis refines this recommendation by identifying short length, sparse orthology, and limited insertion opportunities as confounders: truly dark genes had a median length of 121 amino acids versus 194 for annotation-lag genes, and short genes can be harder both to annotate and to measure with transposon insertions. ₄

Discoveries supports the need for orthogonal perturbation and condition coverage: fitness modules derived by robust independent component analysis (ICA) and DBSCAN were stable at 17–52 modules per organism across 32 bacteria, 94.2% had significantly elevated within-module cofitness, and absolute module-membership thresholding increased within-module mean |r| from 0.047 to 0.34. ₁ However, ortholog scope changed the number of cross-organism module families from 27 to 156, and Module-ICA had <1% KO-level precision despite capturing process-level co-regulation; this refines module-based prioritization by separating gene-level transfer from process-level hypotheses. ₁

The conservation-versus-fitness results refine experimental interpretation of essential dark genes. Essential-core genes numbered 22,751 and were 41.9% enzymes and 13.0% hypothetical, whereas essential-auxiliary genes numbered 3,683 and were 13.4% enzymes and 38.2% hypothetical; essential-unmapped genes were 44.7% hypothetical. ₉ Essential-core genes were the best annotated category, with 87% having known function, while essential-auxiliary and essential-unmapped genes provide a more targeted pool for CRISPRi validation because their essentiality is paired with weaker conservation or annotation evidence. ₉ The essential-unmapped category included divergent ribosomal proteins, translation factors, transposases, and DNA-binding proteins, but these assignments remain hypotheses about recently acquired or highly divergent variants of core functions rather than direct functional proof. ₉

The hypothetical-essential pool is large and splits by predictability. Across 48 organisms, 8,297 hypothetical essential genes make up 20.2% of all essentials. Of these, 3,912 have orthologs and can be predicted by the ortholog-based method, whereas 4,385 are orphans that the method cannot predict. ₁₀ Separately, the report states that 7,084 essential genes have no orthologs in any other Fitness Browser organism and that these are 58.7% hypothetical, compared with 8.2% for universally essential genes. It describes them as the least characterized yet most functionally important genes in each organism. ₁₀ The report does not reconcile the 4,385 hypothetical orphans with the 7,084 total orphan essentials and their 58.7% hypothetical share. These figures appear to come from different tallies or cohorts, so they are preserved as reported and the discrepancy remains unresolved. ₁₀ Module context partly bridges the predictability gap. By finding non-essential orthologs in ICA fitness modules, discoveries generated 1,382 function predictions for hypothetical essential genes across 48 organisms, all backed by cross-organism module-family conservation. These are indirect predictions rather than measurements of the essential genes themselves. ₁ A broader module analysis produced 6,691 function predictions for hypothetical proteins across 32 organisms. ₁₁ These counts support crispri-based prioritization of essential dark genes and mark orphan essentials as the subset where ortholog-based prediction is unavailable and direct perturbation is the main route. _10,1

Functional inference before validation

GapMind identified 1,256 organism–pathway pairs across 44 Fitness Browser-linked species in which nearly complete pathways with low steps_missing_low scores co-occurred with strongly fitness-active dark genes. ₃ The most frequent gaps were fucose utilization in 32 organisms, rhamnose utilization in 31, sorbitol utilization in 30, myoinositol utilization in 28, gluconate utilization in 26, and asparagine biosynthesis in 24. ₃ Marinobacter had 49 gapped pathways, and D. desulfuricans ME-23 and P. stutzeri had 45 each. These organisms also have the largest dark-gene complements and the most specific fitness phenotypes. The report proposes, but does not establish, that their dark genes may encode the missing steps. ₃

These pathway results generate organism-level hypotheses rather than direct gene-to-enzyme assignments. ₃ Domain matching produced 42,239 gene–pathway candidates across 3,186 unique dark genes, including 5,398 high-confidence EC-prefix matches, 4,687 medium-confidence Pfam-family matches, and 32,154 low-confidence keyword matches. ₃ Individual assignments therefore require EC matching, structure prediction, enzymology, or other experimental validation. ₃ This evidence boundary refines metabolic-model-gapfilling and evidence-triangulation-for-functional-annotation by separating pathway plausibility from gene-level functional proof.

Other projects supply validation-ready candidates that fit the same evidence boundary. The annotation-gap analysis identified 44 high-confidence gene–reaction assignments as directly testable by targeted knockouts, such as CRISPRi, on specific carbon sources. It names rxn02185 and rxn03436 across 9 organisms as priority targets. This is a proposed validation, not a reported knockout result. ₁₂ The aromatic-catabolism network analysis found that the DUF-domain proteins aciad3137 (UPF0234) and aciad2176 (DUF2280) show r > 0.98 co-fitness with complex-i genes, making them candidate uncharacterized Complex I accessory factors. Their accessory-factor status remains an experimental validation target, not a demonstrated physical association. ₁₃

Structural evidence offers a separate prioritization route. The AlphaFold multiple-sequence-alignment (MSA) analysis calls the 415,603 core gene clusters with msa_depth < 10 its most scientifically actionable finding. It proposes these paradox proteins as prime candidates for experimental structural characterisation but reports no completed characterization. ₁₄ The analysis also proposes joining the paradox-protein list to Fitness Browser fitness scores, where available, to identify paradox proteins essential for growth in at least one condition, but it reports no results from that join. The overlap between structurally under-sampled core proteins and fitness-based dark-gene candidates is therefore untested. ₁₄

Discoveries supports this boundary through a complementary annotation benchmark: eggNOG had higher COG, KEGG, and Pfam coverage, whereas Bakta had higher GO coverage, product descriptions, and unique UniRef50 links; their union reached 77.3% any-functional-annotation coverage but did not make every product assignment experimentally established. ₁ The finding refines pathway-based prioritization by favoring candidates where independent annotation, fitness, and biochemical evidence converge rather than treating any single database label as proof.

Cross-organism and genomic-context evidence increased prioritization confidence without constituting direct functional validation. ₃ Fitness concordance identified 65 ortholog groups present in at least three organisms whose dark-gene orthologs showed measurable effects under the same condition classes, with several groups reaching concordance of 1.00. ₃ The strongest concordance involved carbon-source genes spanning 3–8 organisms and motility genes in 3 organisms each. The stress-concordant OG05812 carries a Peptidase_M50 (site-2 protease family) domain. The report treats these as hypotheses of conserved, unannotated carbohydrate-metabolism, chemotaxis and regulatory-protease functions, not as assignments. ₃ Cross-species synteny conserved 17,058 of 21,011 tested dark-gene–operon-partner pairs in at least one other organism and 10,150 in at least 3 organisms. ₃ Independent co-fitness analysis found evidence for 2,899 of 32,075 non-essential operon pairs, including 1,129 mutual top-5 pairs, and 998 pairs were double-validated by conserved synteny and strong co-fitness. ₃ These results support cofitness-network-architecture while retaining the distinction between association and experimentally demonstrated function.

The truly dark results support using context as a prioritization signal: among 4,394 such genes with integrated condition analysis, 25.9% showed operon-like cofitness with adjacent genes at |r| ≥ 0.3, and ICA modules linked candidates to functions including phage integrases, metal transporters, chemotaxis systems, and iron regulation. ₄ They also reinforce the need for validation because some strong fitness phenotypes may reflect polar effects on downstream genes. ₄ Discoveries supports the network evidence while calibrating transferability: Pfam domains increased module annotation from 92/1,116 (8.2%) to 890/1,116 (79.7%), and ortholog transfer achieved 95.8% precision and 91.2% coverage for gene-level KO prediction, whereas Module-ICA remained a process-level rather than gene-level predictor. ₁

Conservation and campaign design

Full GTDB r214 pangenome analysis expanded conservation coverage from 32,791 dark genes, or 57.5%, to 37,997, or 66.6%, by querying 93.5M gene-cluster annotations and recovering 5,206 additional dark genes through ortholog-group propagation. ₃ A conservation-weighted covering set selected 42 organisms covering 95.6% of importance-weighted priority across 28,584 high-priority dark genes. ₃ A separate darkness-spectrum analysis classified 57,011 dark genes into T1 Void (4,273), T2 Twilight (12,282), T3 Dusk (16,103), T4 Penumbra (22,500), and T5 Dawn (1,853). ₃

The conservation-by-ignorance classification assigned 6.0% of dark genes to strong testable hypotheses, 52.5% to weak leads, and 41.5% to true knowledge gaps. ₃ The report’s highest-ranked knowledge gaps included COG0468 with 27,427 species across 142 phyla, COG0443 with 27,279 species, and COG0491 with 27,393 species. ₃ These results support genomic-under-representation and pangenome-integration, but broad conservation should be treated as a prioritization signal rather than evidence that a gene has the same function in every lineage. ₃

The Fitness Browser–pangenome analysis supports this calibration while adding a within-species essentiality test: 145,821 linked genes (82.0%) were in core clusters and 32,042 (18.0%) were auxiliary, with 7,574 singleton genes included in the auxiliary category. ₉ Essential genes were more often core than non-essential genes, but the median odds ratio was only 1.56, so essentiality does not identify a uniformly conserved class. ₉ Clade-size and lifestyle stratification indicated that essential-core enrichment was robust across the genomic contexts examined, although clades containing only 2 genomes can have trivially high core fractions and therefore reduced discriminative power. ₉

The truly dark comparison refines this interpretation: persistent hypothetical genes had a core-genome fraction of 43.1% versus 72.7% for annotation-lag genes, ortholog presence of 29.3% versus 63.7%, and median ortholog breadth of 1 versus 4 organisms, with all reported differences strongly significant. ₄ Their shorter length and narrower taxonomic distribution are consistent with biological novelty, but the report notes that these measurements do not by themselves establish a function or recent horizontal gene transfer. ₄

Discoveries supports the need for clade-aware calibration: pangenome openness showed no significant relationship with environment or phylogenetic effects in one analysis (rho=-0.05, p=0.54 and rho=0.03, p=0.73), while variable metabolic pathways strongly predicted openness in another (rho=0.327, p=7.2e-71, and partial rho=0.530, p=2.83e-203 after controlling for genome count). ₁ These results do not replace the existing conservation evidence; they refine campaign design by indicating that inferred novelty and conservation can depend on pathway composition, genome sampling, and the analysis scale.

Environmental validation and its limits

Within-species carrier-versus-non-carrier analyses tested 151 accessory dark-gene clusters across 31 species, and 10 of 137 clusters with environmental-category tests showed significant enrichment at FDR < 0.05. ₃ One of 67 clusters showed significant AlphaEarth embedding divergence. ₃ A pre-registered lab–field mapping found 29 of 47 testable clusters concordant, or 61.7%, with pH-associated genes showing 100% concordance across 4 tests and nitrogen-source genes showing 78% concordance across 9 tests. ₃

NMDC, the National Microbiome Data Collaborative dataset, independently mapped 5 of 6 carrier genera to 47 taxon columns across 6,365 metagenomic samples and confirmed all 4 testable pre-registered abiotic predictions. ₃ Nitrogen-source carriers correlated with total nitrogen at ρ = +0.109 across n = 1,231 samples with FDR = 2.3e-4 and with ammonium nitrogen at ρ = +0.231 across n = 1,230 samples with FDR = 8.0e-16. ₃ pH carriers correlated with pH at ρ = +0.157 across n = 4,366 samples with FDR = 7.4e-25, and anaerobic carriers correlated negatively with dissolved oxygen at ρ = -0.298 across n = 272 samples with FDR = 1.5e-6. ₃

The environmental evidence is supportive but not conclusive. ₃ The one-sided binomial test for 29/47 concordant clusters gave p = 0.072 with Wilson score 95% CI [0.474, 0.742], whereas Fisher’s combined probability across 47 individual tests gave p = 0.031. ₃ NMDC trait-feature analysis confirmed all 7 pre-registered trait-condition predictions at FDR < 10⁻²¹, but 441 of 449 exploratory tests also reached FDR < 0.05, which the report attributes largely to compositional coupling because abundant genera contribute to both carrier abundance and community trait scores. ₃ This limitation connects the framework to confirmatory-exploratory-ecological-association-discordance and taxonomic-resolution-dependent-functional-inference.

Discoveries supports the pre-registered portion but reinforces the caution about exploratory associations: all 4 pre-registered abiotic predictions and all 7 pre-registered trait predictions were confirmed, while the NMDC metabolomics analysis was dominated by one study, with 125/131 samples (95%) from nmdc:sty-11-r2h77870; only approximately 2% of compounds had KEGG IDs, and three amino-acid pathways—cysteine, histidine, and lysine—were untestable because compounds were absent. ₁

Field-derived modules add environmentally framed candidates. The 21 ecological modules contained 239 member genes, 52 of which are unannotated and are proposed as candidates for novel environmental-adaptation functions. This extends the lab–field mapping above with a candidate pool rather than with additional validation. ₁₅

Reannotation and coverage bias

Bakta v1.12.0 with DB v6.0 reclassified 33,105 of 39,532 pangenome-linked dark genes, or 83.7%, as not hypothetical, representing 58.1% of all dark genes. ₃ All 100 top candidates received Bakta product descriptions, and 5 dark genes received AMRFinderPlus annotations. ₃ A further 6,427 genes remained hypothetical in both Fitness Browser and Bakta but gained UniRef50 links for 79.4%, UniParc/UniRef100 links for 69.1%, and RefSeq links for 62.4%. ₃

The truly dark analysis supports the interpretation that most dark-gene labels reflect annotation lag while identifying a residual set: 33,105 genes were reclassified, whereas 6,427 remained hypothetical in both pipelines; only 4.0% of the latter had Pfam hits and 4.6% had KEGG KOs, although 79.4% had UniRef50 links. ₄ These results weaken any interpretation of the 57,011-gene total as a definitive count of true functional darkness because database release and annotation coverage affect whether a gene is labeled unknown. ₃ The residual estimate is itself incomplete: 17,479 dark genes lacked pangenome links, and the report estimates approximately 2,841 additional truly dark genes among them. ₄

Discoveries supports the reannotation result and adds a stronger operational distinction: among 39,532 dark genes with pangenome links, 33,105 (83.7%) were reclassified and 6,427 remained truly dark; truly dark genes were shorter (121 versus 194 aa), less conserved (43% versus 73% core), had fewer orthologs (29% versus 64%), had higher GC deviation (d=0.247), and had 41% of neighbors also hypothetical. ₁ Only 246/6,427 (3.8%) had zero annotation clues, so experimental campaigns should prioritize residual genes with interpretable partial evidence before treating the entire dark-gene set as equally inaccessible.

The report also found strong taxonomic imbalance: 37/48 organisms were Pseudomonadota, and the top 500 candidates included 417 from Gammaproteobacteria, 71 from Alphaproteobacteria, 17 from Deltaproteobacteria, 12 from Bacteroidetes, 10 from Betaproteobacteria, 5 from Firmicutes, 2 from Cyanobacteria, and 0 from Archaea, Actinobacteria, or Epsilonproteobacteria. ₃ The framework therefore supports structural-annotation-gap and genetic-perturbation-coverage-bias by making candidate rankings conditional on both annotation resources and perturbation availability. ₃

Pangenome linkage introduces additional coverage constraints: four Fitness Browser organisms—Cola, Kang, Magneto, and SB2B—were unmatched because their species had too few genomes for GTDB pangenome construction; Dyella79 was excluded because a locus-tag mismatch produced a 0% join rate, and ten organisms were excluded from Phase 2 because they had <90% DIAMOND coverage. ₉ The main Escherichia coli clade was absent because it contained too many genomes, while Keio mapped to a small s__Escherichia_coli_E clade at only 26.1% coverage. ₉ These limitations refine the interpretation of conservation-weighted rankings: missing or weakly sampled clades can affect both conservation labels and the apparent experimental priority of unknown genes.

Discoveries further refines this coverage warning: only 6.8% of species (1,872/27,690) had sufficient AlphaEarth coverage for niche-breadth analyses, and 3,838/83,287 genomes had NaN dimensions; 36.6% of genomes clustered at coordinates with more than 50 genomes of more than 10 species. ₁ Environmental and geographic scores should therefore be reported with coverage filters and treated as conditional evidence, not as universally available ranking dimensions.

Route-based campaign design and per-organism action plans

The report turns the analysis into an experimental campaign built on two complementary prioritization routes. One route is optimized for genes with the strongest existing evidence and the other for discovering completely unknown functions, and both converge on a tractable set of organisms. ₃ Route A (evidence-weighted, NB05–NB09) scores each dark gene across 6 dimensions using data from the 48 Fitness Browser organisms: fitness importance 0.25, conservation 0.20, inference 0.20, pangenome 0.15, biogeographic 0.10 and tractability 0.10. It is best suited to targeted experiments where the condition to test is already known. ₃ Route B (conservation-weighted, NB11 and NB11b) queries the full 27,690-species gtdb r214 pangenome to measure the taxonomic breadth of each dark-gene OG (species range 1–27,482). It classifies OGs by taxonomic tier (kingdom through species) and by hypothesis status, then ranks them by importance = conservation × ignorance, for discovery experiments on broadly conserved unknowns. ₃ This supports the complementary-routes reading from discoveries above, and it states the route weights and data scope explicitly. _3,1

Route A’s top fitness-active candidates are genes where fitness data, module co-regulation, domain annotations and biogeographic signals converge on a testable hypothesis. ₃ The first-ranked candidate is P. putida N2C3 AO356_11255, the only accessory gene in the top 10. It has a score of 0.715, a D-alanyl-D-alanine carboxypeptidase prediction (EC 3.4.16.4) and nitrogen fit=3.4, supported by a module prediction, an EamA domain, lab-field OR=44 and an NMDC nitrogen correlation (see the FDR caveat above and lab-field-fitness-concordance). ₃ shewanella-oneidensis MR-1 supplies the next four candidates:

• 202463 (0.698; PF01145 stress-responsive membrane protein; stress fit=6.4; YGGT domain and a co-regulated module);
• 199738 (0.698; K03306 family; nitrogen fit=5.5);
• 203545 (0.694; K03306 family with a DUF4124 domain; nitrogen fit=4.0);
• 202450 (0.693; K03306 family with a Gly_transporter domain; nitrogen fit=3.9). ₃ The report reads the K03306 trio as paralogous members of a conserved nitrogen-responsive system. It takes 203545’s different domain within the same module as a sign of subfunctionalization, and it reads 202463 and 203247 as a stress-responsive membrane-protein family (YGGT/GBBH-like). These are hypotheses from module and domain evidence, not demonstrated functions. They support the MR-1 screen designs described above. ₃

Among the most widespread top-500 dark gene clusters (present in 15+ pangenome species), the most common domain families are TauE (sulfonate export, 7 clusters), DUF444 (5 clusters), Cu-oxidase_4 (5 clusters), EamA (4 clusters, including the #1 candidate) and DUF484/DUF971/DUF934 (3+ clusters each). ₃ This ties the top fitness-active candidate to a recurring, widespread domain family and connects candidate selection to core-gene-annotation-paradox. A recurring domain family is a prioritization cue, not a shared-function assignment. ₃

Incorporating additional evidence improved scores for 10,757 fitness-active and 4,028 essential dark genes, and some candidates rose significantly in rank. ₃ A cross-check of the top 100 candidates against prior module-based predictions found that 85/100 had existing predictions, all agreeing with the project’s independent inference. ₃ The report calls this strong orthogonal validation. It is concordance between two computational inference routes rather than experimental confirmation, so it supports the hypothesis quality of the candidates without establishing their functions (see evidence-triangulation-for-functional-annotation). ₃

The roadmap is a projection built on existing Fitness Browser RB-TnSeq libraries. It ranks all organism–condition pairs by the number of top-500 dark-gene candidates they would address and then applies greedy set-cover optimization, so its coverage figures describe planned experiments, not completed ones. ₃ In a later iteration of the evidence-weighted roadmap, MR-1 ranks first with 61 candidates (47 fitness-active + 14 essential, 83 synteny-validated pairs and 31 cofit-validated pairs). P. putida N2C3 (35 candidates) and P. putida (24 candidates) follow, and just 4 experiments achieve 45% coverage of the top 300 candidates. ₃ This refines the earlier top-500 projection, but the denominator changed from the top 500 to the top 300, so the two coverage percentages are not directly comparable. ₃

A greedy weighted set cover over the 16,488 scored dark genes selects 42 organisms (from 28 genera) that cover 95% of total priority value. MR-1 ranks first, contributing 587 genes, and 32 organisms suffice for 80% coverage. ₃ The algorithm optimizes composite priority score × tractability × phylogenetic diversity. ₃ The resulting per-organism action plans split these genes into 14,450 hypothesis-bearing genes, which have specific condition recommendations from fitness data, neighbor context or module predictions, and 2,038 darkest genes that require broad phenotypic screens. A further 8,900 essential genes are flagged for CRISPRi-based approaches. ₃ The two routes converge on the same organism count. Route A selects 42 organisms (28 genera) covering 95% of composite priority, and Route B selects 42 (28 genera) covering 95.6% of importance-weighted priority. The routes share 39 organisms: Route A uniquely selects 3 (Dda3937, Ddia6719, Miya) and Route B uniquely selects 3 (Kang, Methanococcus_JJ, Methanococcus_S2). ₃ The priority totals differ between routes: composite priority is computed over 16,488 scored genes, whereas importance-weighted priority is computed over 28,584 high-priority genes. The similar percentages therefore do not mean that the routes cover the same genes. ₃ The report summarizes the two routes as producing different orderings that reflect different experimental strategies. It also offers an extended tractable list of 73 organisms (48 Fitness Browser plus 25 literature-curated from 3 additional phyla) as a way to expand coverage beyond the Proteobacteria-dominated Fitness Browser. ₃

The Route A per-organism plan places these organisms after MR-1:

• Pseudomonas fluorescens N1B4 ranks second (624 genes, tractability 0.7). It contributes the most additional uncovered genes after MR-1, particularly in carbon/nitrogen source panels (215 genes) and membrane stress (77 genes).
• sinorhizobium-meliloti ranks third (570 genes, tractability 0.6). It covers a distinct Alphaproteobacteria lineage and has 316 T4 Penumbra and 90 T5 Dawn genes.
• escherichia-coli K-12 (Keio) ranks fourth (368 genes, tractability 0.9). It has the highest tractability, with 259 hypothesis-bearing genes immediately testable and 205 essential genes ideal for CRISPRi.
• klebsiella-oxytoca ranks fifth (396 genes, tractability 0.65). It covers Enterobacteriaceae genes not found in E. coli, including top essential candidate BWI76_RS08540 (score 0.865). ₃ This ordering differs from the candidate-count roadmap, which places P. putida N2C3 second; see Tensions. ₃

Route B’s conservation-weighted set cover (NB11) selects 42 organisms (28 genera) covering 95.6% of total importance-weighted priority across 28,584 high-priority dark genes. It chooses organisms by sum(importance) × tractability × phylo_bonus, where phylo_bonus penalizes genus-redundant selections. ₃ The report gives the leading organisms’ counts in two different units. One passage reports them as genes: sinorhizobium-meliloti first (1,630 genes, 195 kingdom-level gaps), followed by pseudomonas-putida (1,043 genes, 172 kingdom gaps), S. oneidensis MR-1 (805 genes, 105 kingdom gaps) and bacteroides-thetaiotaomicron (1,382 genes, 267 kingdom gaps, described as the highest raw count and deprioritized by 0.3 tractability). ₃ The per-organism Route B plan reports the same values as OGs:

• S. meliloti first (1,630 OGs, 195 kingdom gaps, tractability 0.6);
• P. putida second (1,043 OGs, 172 kingdom gaps, tractability 0.8);
• MR-1 third (805 OGs, 105 kingdom gaps, tractability 0.8);
• B. thetaiotaomicron fourth (1,382 OGs, 267 kingdom gaps, tractability 0.3);
• klebsiella-michiganensis fifth (568 OGs, 65 kingdom gaps, tractability 0.65). ₃ The report does not say whether genes or OGs is the correct unit, so both are preserved as reported. ₃ The report’s superlatives also conflict with its own numbers. It says S. meliloti was selected first because its dark genes span the most kingdom-level true knowledge gaps, yet it lists 195 kingdom gaps for S. meliloti and 267 for B. thetaiotaomicron. It also says B. thetaiotaomicron has the highest raw OG count of any organism, yet it lists 1,382 OGs for B. thetaiotaomicron and 1,630 for S. meliloti. The report does not explain either discrepancy, so neither superlative should be read as an established maximum. ₃ Route A ranks MR-1 first, for deep condition coverage and many converging evidence lines, whereas Route B ranks S. meliloti first. MR-1 falls to third in Route B because that route weights conservation breadth over condition-specific evidence depth. This is consistent with fitness-condition-coverage-prioritization-bias shaping the Route A lead. ₃ The report calls B. thetaiotaomicron critical for non-Proteobacteria (Bacteroidota) coverage but deprioritizes it for low tractability (0.3). Tractability weighting can therefore push taxonomically valuable organisms down the list. ₃ The first 10 Route B organisms cover 31% of importance-weighted priority (9,014 genes across 10 genera). ₃

The Route A per-organism plan explains MR-1’s lead. MR-1 has 587 genes, tractability 0.8, 121 historical conditions and a dark-gene complement spanning all tiers, including 172 T5 Dawn and 257 T4 Penumbra genes. Of its dark genes, 544 have specific condition recommendations (stress, nitrogen, carbon), and 25/100 top fitness-active candidates reside in MR-1. ₃ Route A’s conservation assessment is limited to the 48 Fitness Browser organisms because the NB09 evidence-weighted approach relies on Fitness Browser data. ₃ For labs focused on novel function discovery rather than hypothesis testing, the report proposes the Route B ordering. That ordering starts with S. meliloti (195 kingdom-level gaps), P. putida (172 kingdom gaps) and MR-1 (105 kingdom gaps). The report recommends broad phenotypic screens under diverse condition panels because the target genes are true knowledge gaps without condition predictions. It says this ordering maximizes the probability of uncovering completely new biology, but that benefit is a prospective claim, not a measured yield. ₃

Route B targets the most broadly conserved genes with zero functional evidence, where the report expects experimental characterization to produce the most novel biological insight. ₃ Each dark gene is independently assigned one of three hypothesis-status tiers:

• strong testable hypothesis (6.0%): a module prediction with EC number, high cross-organism concordance, a high-confidence GapMind match, or a named domain + strong fitness;
• weak lead (52.5%): a DUF-only domain, a bare module prediction, a medium/low GapMind match, strong fitness without annotation, or a named domain without fitness;
• true knowledge gap (41.5%): zero functional evidence of any kind. ₃ The importance score conservation × ignorance ranks all 11,774 dark gene OGs. Kingdom-level true knowledge gaps score highest and species-specific strong hypotheses score lowest. ₃ More precisely, the importance score is (tier-adjusted conservation + log₂-scaled species fraction) × an ignorance multiplier. The full ranking of 11,774 OGs is released as data/og_importance_ranked.tsv. ₃ The top-ranked OGs named in this ranking are cog0468 (27,427 species, 142 phyla, true knowledge gap), cog0443 (27,279 species, true knowledge gap) and COG0172 (27,431 species, weak lead). The report describes them as among the most universally conserved genes in bacteria yet functionally uncharacterized. ₃ This refines the knowledge-gap list above, which names COG0491 (27,393 species) alongside COG0468 and COG0443, by adding COG0172 as a highly ranked weak lead. The source passages do not explain the difference in the third-named OG. ₃ Breadth of conservation remains a ranking signal rather than evidence of a conserved function across lineages, a point that bears on research-attention-inequality. ₃

The darkness-tier census assigns each of the 57,011 dark genes a tier based on 6 binary evidence flags: domain annotation, ortholog group, function prediction, co-fitness partner, fitness/essentiality phenotype and pangenome context. Tiers range from T1 Void (4,273 genes, 0 evidence lines, truly unknown) to T5 Dawn (1,853 genes, 5–6 evidence lines, nearly characterized). ₃ Because the tiers count independent evidence flags, they support the split between routes: the darkest tiers feed broad screens, while higher tiers carry condition-specific hypotheses. ₃ The report states that only 7.5% of dark genes (4,273 T1 Void) are truly unknown with zero evidence, whereas 39.5% are T4 Penumbra genes with 3–4 converging lines of evidence. It proposes tier-matched follow-up: T5 Dawn genes need confirmation, T4 Penumbra genes need targeted experiments and T1 Void genes need broad screens. This supports the tier percentages reported by discoveries above, which the source report itself also states. ₃

Per-organism experimental plans match each evidence type to an assay. They specify CRISPRi knockdown for essential genes in tractable organisms, condition-specific TnSeq for fitness-active genes and a broad phenotypic screen for true knowledge gaps, with top conditions derived from fitness data or keyword inference. ₃ The 14,450 hypothesis-bearing genes across all 42 organisms each have at least one condition recommendation from fitness data, module co-regulation or gene-neighborhood context. For these genes the proposed experiment is targeted RB-TnSeq or CRISPRi under the predicted condition. For example, MR-1:202463 (YGGT domain) should be tested under oxidative, osmotic and heat stress, and the K03306 paralog trio (199738/203545/202450) under nitrogen limitation with single- and double-mutant comparisons. ₃

The action plan also reports a null result for hypothesis generation. Across all 42 organisms, 2,038 genes are T1 Void or T2 Twilight genes that have detectable fitness phenotypes but no converging evidence pointing to a specific condition. ₃ For these genes the report recommends a broad phenotypic screen: RB-TnSeq across a diverse panel following the standard Fitness Browser protocol of ~50-100 conditions. E. coli K-12 alone has 109 such genes. ₃ A measured phenotype therefore does not by itself yield a targeted experiment, which supports keeping condition-specific-fitness evidence separate from functional-hypothesis evidence. ₃

The covering set contains 8,900 essential genes, which cannot be studied by transposon knockout. For these the report recommends crispri knockdown, using Mobile-CRISPRi for non-model organisms (Peters et al. 2019), with growth curves under standard and stress conditions. ₃ All top-50 essential candidates have high-confidence functional hypotheses derived from operon context. Each comes with a specific CRISPRi design naming the target organism, sgRNA target, growth condition, expected phenotype and validation strategy. ₃ For example, the #1 essential candidate, Keio:14796 (YbeY domain), is predicted to function in ion transport because it shares an operon with an annotated ion-transport gene. The proposed test is CRISPRi knockdown in E. coli Keio grown on varied carbon/nitrogen sources, monitoring OD600 for growth defects. ₃ The high-confidence label reflects operon-context support, and these assignments remain predictions until the knockdowns are run. ₃

The ten highest-ranked essential dark-gene candidates, all labeled high confidence, are listed below as organism and locus, then score, domain and operon context:

1. E. coli Keio 14796: 0.875; YbeY, TIGR00043; ion transport + NTP hydrolase.
2. Shewanella MR-1 200382: 0.874; RimP_N, DUF150_C; tRNA-Met + NusA.
3. K. oxytoca BWI76_RS08540: 0.865; OmpA, TIGR02802; TolB + CpoB (cell division).
4. P. putida N2C3 AO356_29395: 0.838; Peptidase_M20; ABC transporter + peptidase.
5. Shewanella MR-1 201473: 0.835; EarP, TIGR03837; EF-P + flavodoxin.
6. E. coli Keio 14768: 0.833; DUF493; lipoate biosynthesis + PBP5.
7. Shewanella MR-1 200359: 0.833; YbeY, TIGR00043; CorC + PhoH.
8. P. putida PP_1910: 0.828; YceD; 50S ribosomal L32 + PlsX.
9. P. putida PP_5002: 0.828; GBBH-like_N; HslVU protease + PhaC1.
10. E. coli Keio 11474: 0.825; DUF4109; tRNA-Met + translation. ₃ The YbeY/TIGR00043 domain recurs in Keio 14796 and MR-1 200359. The operon-context functions remain predictions, and the essential-gene scores come from a ranking separate from the fitness-active Route A scores. ₃

The report’s annotations for the top three essential candidates are also predictions rather than demonstrated functions:

• E. coli K-12 14796 (score 0.875, YbeY) is conserved in 5+ phyla, sits in a synteny-confirmed operon and has predicted ion transport.
• S. oneidensis MR-1 200382 (0.874, RimP_N/DUF150_C) is a predicted ribosome assembly factor with a conserved neighborhood.
• K. oxytoca BWI76_RS08540 (0.865, OmpA/TIGR02802) is predicted to function in cell division and is conserved across Enterobacteriaceae. ₃ The report attributes their high scores to several converging evidence lines: detectable domains, widely conserved ortholog groups, conserved gene neighborhoods with annotated partners and organisms amenable to CRISPRi. These supply functional clues and experimental tractability, not experimental confirmation. ₃

Essential genes needed their own ranking. None appear in the NB05 top 100 because essential genes have zero rows in genefitness. The fitness-importance dimension therefore scores them at the essentiality-bonus floor, and the tractability dimension penalizes them as non-knockable. ₃ Essential genes have no fitness scores beyond the essentiality call, so this null result conflicts with the report’s earlier assertion that every catalog gene has measured fitness phenotypes. ₃ The report describes essential dark genes as the majority (55%) of phenotype-bearing dark genes and says fitness-based scoring systematically misses them. ₃ A separate NB07 prioritization therefore combines gene-neighborhood context, domain structure, ortholog conservation, phylogenetic breadth and CRISPRi tractability. It produces 50 ranked essential candidates amenable to CRISPRi knockdown. ₃ The top 50 essential candidates score 0.740–0.875, span 15 organisms and all carry high-confidence hypotheses from neighbor context and domain annotations. The report attributes their higher scores, relative to the fitness-active top 100 (0.624–0.715), to stronger conservation and neighborhood signals. The two sets were, however, scored by different approaches. ₃

The campaign’s first three proposed experiments are:

1. An MR-1 stress screen: RB-TnSeq under 5 stress conditions (oxidative, osmotic, metal, heat, pH), addressing 161 hypothesis-bearing dark genes, including the #2 overall candidate (202463, fit=6.4 under stress).
2. An MR-1 nitrogen screen: RB-TnSeq under nitrogen limitation and amino acid supplements, addressing the K03306 paralog family (3 genes, fits 3.9–5.5) and 74 total dark genes.
3. An E. coli K-12 CRISPRi experiment: knockdown of the top 20 essential dark genes with growth curves, using the highest-tractability organism (0.9) and its mature CRISPRi tools, with top target Keio:14796 (YbeY domain, score 0.875). ₃ These are proposals, not completed screens. Their gene counts differ from the top-500 set-cover projection above, in which the MR-1 stress screen addresses 57 genes and the nitrogen screen 31, and the third slot there is an MR-1 carbon screen. The campaign appears to count hypothesis-bearing genes in the covering set rather than top-500 candidates, but the report does not reconcile the two tallies. ₃

The report also projects that adding bacillus-subtilis and S. coelicolor first would fill the two largest phylogenetic gaps, Firmicutes depth and Actinobacteria absence. Together these additions would enable cross-phylum testing of the ~100 “universal” dark gene families that can currently be studied only in Proteobacteria. This is a prospective claim. ₃ The extended covering set (NB11c) tests that gap computationally. It uses a curated list of 73 organisms: the 48 Fitness Browser organisms, which are 77% Pseudomonadota (37/48), plus 25 literature-curated organisms with TnSeq/CRISPRi resources from Bacillota, Actinomycetota and Campylobacterota. The list was assembled to address the Proteobacteria-heavy Fitness Browser sampling bias. ₃ A Spark query mapped genus-level OG membership for 25 non-FB organisms (24 genera, 53,970 genus-OG pairs in 2.9 min). Running the conservation-weighted cover with all 73 candidates yields a 50-organism set covering 98.7% of OGs across 6 phyla, versus 41 organisms, 100% and 4 phyla for FB-only candidates. ₃ The 98.7% is an OG-coverage percentage, not an importance-weighted priority percentage like Route B’s 95.6%. ₃ The cover selects 16 non-FB organisms, led by these:

• pseudomonas-aeruginosa PAO1 (#1, 3,713 OGs, tractability 0.8);
• V. cholerae N16961 (1,049 OGs);
• B. cenocepacia K56-2 (956 OGs);
• mycobacterium-tuberculosis H37Rv (#6, 131 new OGs from Actinomycetota);
• C. jejuni (#48, Campylobacterota). ₃ Coverage at N=5 organisms reaches 57.7%, versus 38.1% FB-only (+19.6%). ₃ The report’s own limitations qualify these gains. Non-FB OG coverage is estimated at the genus level, counting an OG as present if any species in the genus has it, which overestimates individual-organism coverage. Non-FB organisms also lack Fitness Browser condition profiling, so their dark genes cannot be assigned to condition-specific experiments and can only enter broad phenotypic screens. ₃ The added phyla therefore broaden discovery-screen scope rather than targeted-test scope, and the OG coverage figures for non-FB organisms are upper estimates. ₃ The Bacillota organisms B. subtilis and staphylococcus-aureus were not selected because their OGs are subsets of coverage already provided by Pseudomonadota. They remain valuable for studying genes in native Gram-positive genomic context. ₃ This creates a tension within the report. The recommendation to add B. subtilis first rests on phylogenetic-gap reasoning, while OG set cover does not select it, because the two criteria answer different questions: cross-phylum replication versus unique-OG coverage. The tension bears on genetic-perturbation-coverage-bias. ₃

The report’s figures document these steps:

• fig13_score_components.png: scoring-dimension distributions for all dark genes;
• fig14_top20_dossiers.png: score components for the top 20 candidates (stacked bar);
• fig15_organism_distribution.png: the top 100 candidates by organism and condition class;
• fig19_essential_scores.png: essential-dark-gene score distributions across 5 dimensions;
• fig20_essential_top20.png: score breakdowns for the top 20 essential candidates;
• fig24_improved_experimental_roadmap.png: the evidence-weighted roadmap’s coverage curve, organism heatmap and priority breakdown;
• fig25_darkness_spectrum.png: darkness-tier distribution, evidence-flag combinations and composite score by tier;
• fig26_covering_set.png: set-cover diminishing returns, per-organism tier breakdown and genus-diversity progression;
• fig27_action_plan.png: the organism action-plan heatmap, condition recommendations and tractability versus gene count;
• fig29_robust_ranks.png: rank-range distributions, min/max rank scatter and always-top-50 candidates by organism;
• fig36_covering_set_curve.png: cumulative coverage and marginal genes in the conservation-weighted set cover;
• fig37_experiment_plan_heatmap.png: per-organism dark-gene coverage by conservation tier for the top 20 organisms;
• fig39_extended_covering_set.png: the FB-only versus 73-organism covering-set comparison, including marginal coverage, phylum diversity and kingdom-level gap coverage. ₃

Tensions

The framework simultaneously prioritizes genes with strong measured phenotypes and genes that are conserved or environmentally structured, but these evidence types do not guarantee the same ranking. ₃ Conservation-dominant or drop-tractability settings retained only 64% of the original fitness-active top 50, while essential-gene rankings changed substantially when tractability or neighbor context was removed. ₃ Environmental associations also support hypotheses while remaining vulnerable to sparse metadata, genus-level matching, common-genus effects, and compositional coupling. ₃

The conservation-versus-fitness evidence adds a related tension rather than resolving it: essential genes were enriched in the core genome, but the median odds ratio was 1.56 and essentiality was measured under a single growth condition represented by RB-TnSeq library construction. ₉ Thus, the modest core enrichment does not establish that genes important under stress or other unmeasured conditions will be conserved in the same way. ₉ The essential-gene definition is also an upper bound because missing insertions can reflect short genes, low-complexity regions, or scaffold edges; essential genes were slightly shorter on average, consistent with insertion bias. ₉

A second tension concerns module and neighborhood evidence: 6,142 dark genes belonged to ICA fitness modules, 30,190 dark genes shared a predicted operon with an annotated gene, and 97.2% had at least one annotated neighbor within a five-gene window, but module and neighborhood predictions are guilt-by-association inferences rather than direct experimental validation. ₃ The report further cautions that the 97.2% neighbor rate is expected given a 75% genome-wide annotation rate. ₃ Discoveries adds a methodological tension: ortholog transfer achieved 95.8% precision and 91.2% coverage for gene-level KO prediction, but Module-ICA had <1% KO-level precision while capturing process-level co-regulation. ₁ These should remain distinct evidence products rather than being combined into one confidence score without calibration.

A third, subset-specific tension concerns stress interpretation. The overall dark-gene analysis reports stress conditions as producing the largest concentration of strong dark-gene phenotypes. ₃ In contrast, the truly dark analysis reports stress proportions of 28.7% versus 43.2% for annotation-lag genes among strong-phenotype genes, with OR = 0.53 and p < 0.001, while separately reporting stress at 43.3% versus 54.7% in another comparison. ₄ The results should therefore not be collapsed: stress may be prominent among dark genes overall while being less characteristic of the residual genes that resist reannotation.

A fourth tension concerns pangenome openness and environmental inference. One discoveries analysis found no significant relationship between openness and environment or phylogenetic effects (rho=-0.05, p=0.54 and rho=0.03, p=0.73), whereas another found strong associations between variable metabolic pathways and openness (rho=0.327, p=7.2e-71, and partial rho=0.530, p=2.83e-203 after controlling for genome count). ₁ These results contradict one another at the level of the tested predictors, not necessarily at the level of the prioritization framework; the discrepancy should be retained until feature definitions, species coverage, genome-count control, and analysis cohorts are reconciled.

A fifth tension lies within the functional dark matter report itself, whose counts and ranks differ across roadmap iterations. The MR-1 stress screen addresses 57 top-500 genes in one projection and 161 hypothesis-bearing genes in the campaign, and the nitrogen screen addresses 31 versus 74. P. putida N2C3 is second in the candidate-count roadmap (35 candidates) and is described as the second-highest-value organism by top-500 count. In contrast, P. fluorescens N1B4 is second in the Route A per-organism plan (624 genes). Gene 202463 (fit=6.4) is described as having the highest fitness magnitude among top candidates, whereas 202474 (|fit|=7.1) and 202608 (|fit|=7.3) are reported as the highest fitness magnitudes in the top 100. The Route B leaders’ counts are reported as genes in one passage and as OGs in another, with identical values: S. meliloti 1,630, P. putida 1,043, MR-1 805 and B. thetaiotaomicron 1,382. The report’s statements that S. meliloti spans the most kingdom-level gaps and that B. thetaiotaomicron has the highest raw OG count are not consistent with the counts it lists (195 versus 267 kingdom gaps; 1,630 versus 1,382). These differences may reflect different candidate pools, scopes, units or weighting, but the report does not reconcile them, so they are preserved as reported. ₃

A sixth tension concerns the scope of the co-fitness and re-scoring evidence. One section describes testing 32,075 pairs and calls co-fitness-confirmed operons the strongest functional inference for essential genes that lack direct fitness data. The report’s Finding 12, by contrast, specifies 32,075 non-essential operon pairs and states that co-fitness is unavailable for essential genes. ₃ The same section reports that re-scoring all 17,344 fitness-active and 9,557 essential dark genes with synteny and co-fitness evidence produced 300 improved candidates (200 fitness-active + 100 essential). Finding 1, however, defines the 17,344 as including the 7,787 strong-fitness-effect genes and the 9,557 essentials. ₃ The report therefore conflicts with itself on the tested-pair scope, the population counts and whether co-fitness applies to essential genes, and these statements are preserved as reported. ₃

Open Directions

• Re-run the NMDC carrier–trait analysis with full sample-label permutation tests to determine how much of the 441/449 exploratory-test signal remains after controlling for compositional coupling. ₃
• Process equivalent annotated-accessory-gene controls through the complete biogeographic pipeline to test whether dark-gene environmental enrichments exceed the baseline produced by annotation and sampling structure. ₃
• Use CRISPRi in the highest-ranked essential candidates and measure growth under standard, stress, nitrogen-source, and carbon-source conditions to test whether neighborhood- and domain-based predictions produce reproducible phenotypes. ₃
• Compare essential-core, essential-auxiliary, and essential-unmapped genes across matched RB-TnSeq conditions and pangenome clade sizes to test whether conservation predicts condition-specific fitness beyond binary essentiality. ₉
• Combine EC matching, structure prediction, purified-protein enzymology, and pathway growth assays for the GapMind candidates to resolve organism-level pathway co-occurrence into direct gene-to-reaction assignments. ₃
• Repeat the six-axis prioritization with expanded Fitness Browser condition coverage and more evenly distributed taxa to quantify how much candidate rank is driven by assay depth and Pseudomonadota representation. ₃
• Compare conserved-synteny, co-fitness, and CRISPRi perturbation results for the 998 double-validated operon pairs to test whether network evidence predicts causal functional coupling. ₃
• Extend pangenome linkage to the 17,479 unlinked dark genes and test whether the estimated approximately 2,841 additional truly dark genes show the same clue, essentiality, and genomic-context distributions as the linked set. ₄
• Test whether the 3,683 essential-auxiliary genes compensate for missing core functions by combining ortholog-family analysis, pathway reconstruction, and targeted CRISPRi under the library condition and stress conditions. ₉
• Apply structure prediction and Foldseek searches to the 100 persistent truly dark candidates, then test the highest-ranked motility, nitrogen-source, stress, and transporter-associated hypotheses experimentally. ₄
• Reconcile the conflicting openness results by rerunning both predictor sets on the same species cohort with genome-count adjustment, explicit pathway definitions, and matched environment and phylogeny covariates. ₁
• Benchmark candidate rankings with Bakta–eggNOG union annotations, Pfam evidence, ortholog transfer, and independent CRISPRi outcomes to quantify whether multi-pipeline annotation improves experimental yield rather than merely reducing the dark-gene label. ₁
• Expand AlphaEarth-linked environmental validation beyond the 6.8% of species with sufficient coverage and test whether carrier–environment concordance persists after excluding dense coordinate clusters and correcting for study and sampling structure. ₁
• Join the 415,603 low-depth core clusters (msa_depth < 10) to Fitness Browser fitness and essentiality calls to test whether structurally under-sampled core proteins overlap the fitness-ranked dark-gene candidates. The report proposes this join but does not run it. ₁₄
• Reconcile the hypothetical-orphan essential counts (4,385 versus the 7,084 orphan essentials reported as 58.7% hypothetical) by recomputing both tallies on one explicitly defined cohort of organisms and ortholog calls. ₁₀
• Recompute MR-1 screen coverage, organism ranks and the Route B gene-versus-OG and kingdom-gap counts on one explicitly defined candidate pool. This would reconcile the 57 versus 161 stress-screen and 31 versus 74 nitrogen-screen tallies across roadmap iterations, as well as the conflicting superlatives for S. meliloti and B. thetaiotaomicron. ₃
• Rerun the extended covering set using species- or strain-level OG membership rather than genus-level membership, and add a cross-phylum replication objective alongside unique-OG coverage. This would test whether adding B. subtilis and S. coelicolor enables tests of the ~100 “universal” dark gene families that OG set cover alone does not reward. ₃
• Run the proposed E. coli K-12 CRISPRi knockdowns of the top 20 essential dark genes and score whether operon-context predictions such as Keio:14796 ion transport produce the predicted condition-specific growth defects. ₃
• Re-derive the co-fitness evidence for the 300 re-scored candidates on an explicitly defined pair set that separates non-essential operon pairs from essential-gene pairs. This would show whether any of the 100 essential re-scored candidates actually received co-fitness support. ₃
• Run the proposed Ni, Co, Cu, Zn, W and Mo panel on Methanococcus maripaludis S2 and JJ, which each carry 54 novel metal candidates (reported as genes in the text and OGs in the table) but only iron fitness data. Then test whether the resulting phenotypes fall on the Methanococcus_S2 and Methanococcus_JJ genes in the truly dark top 100. _5,4
• Join the 129,823 PaperBLAST–Fitness Browser VIMSS cross-references to the Route A and truly dark candidate lists to test whether literature coverage predicts which candidates already have published phenotypes. _7,3,4
• Check whether OG01383 in the metal specificity analysis and OG01383 in the functional dark matter roadmap denote the same ortholog group. If they do, test the metal-specific and pH-associated phenotypes in the same strains. _6,3