Contributions
The project AlphaFold MSA Depth as a Lens on the Bacterial Annotation Gap evaluated AlphaFold multiple-sequence-alignment (MSA) depth as a proxy for functional annotation richness across bacterial pangenome gene clusters 1. It joined kbase_ke_pangenome, bakta_annotations, interproscan_domains and kescience_alphafold from a static version-6 KBase Data Lakehouse AlphaFold snapshot 1.
Coverage. Of 132,531,501 gene clusters, 38,804,903 (29.3%) had a real UniProt accession and 38,051,842 (28.7%) bridged to AlphaFold MSA depths 1. The report’s coverage figures are internally inconsistent 1. It gives 70.7% as excluded in one section and 71% as lacking AlphaFold entries in another 1. The 70.7% figure is not the complement of the 28.7% bridge rate 1. The project noted that the analysed subset over-represents well-characterised taxa such as E. coli, Pseudomonas and Bacillus 1. A larger annotation gap among the excluded clusters remains a hypothesis that was not measured 1.
Core versus accessory clusters. The project found median MSA depths of 15,308 for core clusters, 5,299 for auxiliary+singleton clusters and 5,527 for auxiliary non-singleton clusters 1. That makes core 2.89× and 2.77× higher, respectively 1. Hypothetical-protein rates fell from 13.8% (auxiliary+singleton) to 11.6% (auxiliary non-singleton) to 3.8% (core) 1. Chi-square tests gave χ² > 500,000 and p ≈ 0, with odds ratios of 0.25 and 0.31 for core versus each accessory class, linking pangenome structure to annotation coverage (pangenome-integration) 1.
MSA depth and domain annotation. Across 38,051,842 cluster–UniProt pairs, the project found a positive Spearman ρ = 0.7563 between MSA depth and domain-hit count 1. The correlation was not stratified by subgroup and is not a causal test 1. MSA depth was looked up only for each cluster’s representative sequence, so within-cluster diversity was ignored and the representative may have higher or lower depth than typical members 1. In core clusters, mean domain hits rose from 0.59 at MSA depth < 10 to 10.83 at depth ≥ 10,000 1. Mean distinct InterPro (IPR) families rose from 0.059 to 4.601 over the same range 1. The monotone relationship held within each pangenome class, but the report gave no class-specific coefficients 1. InterProScan annotated 111,035,431 clusters (83.8%), averaging 7.5 hits and 3.3 distinct IPR families per annotated cluster 1. The report interpreted this as sequence-profile annotation reaching further than structural homology 1.
Paradox proteins. The project identified 415,603 distinct core clusters with MSA depth < 10 across 14,768 species clades 1. They had a mean MSA depth of 4.57 and a median of 4.0 1. Of these clusters, 68.9% were hypothetical, 0.033% carried Enzyme Commission (EC) annotations and 0.083% were mapped to Kyoto Encyclopedia of Genes and Genomes (KEGG) entries 1. A separate table in the report counts 415,733 very-low-depth core clusters, and the report does not reconcile the two figures 1. The report also contradicts itself by stating that the bottom 10% of core genes have MSA depth ≥ 334 while its own table lists 415,733 core clusters below depth 10 1.
The report presents these proteins as candidates for experimental structural characterisation 1. Its claims of purifying selection and unprecedented folds are interpretations rather than measurements 1. It explained the apparent tension between low overall core hypothetical rates and this subset as two annotation-gap layers (core-gene-annotation-paradox), one driven by MSA depth and one driven by pangenome class 1. The mechanisms it offers for the class-driven gap were not tested 1.
The project proposed within-class stratification, an ESMFold comparison, mapping paradox clusters onto the GTDB (Genome Taxonomy Database) phylogeny, and joining paradox proteins to Fitness Browser scores, but reports no results for any of them 1.