This is an AI-generated wiki. A pipeline built every page on this site from the BERIL Research Observatory’s corpus of AI-conducted microbial-biology research reports.
It is partially reviewed. Scientists have checked some of the research underneath these pages, but no one reviewed the pages themselves before they went up, and nothing marks which claims fall on which side of that line. Do not assume anyone checked a particular claim. Read the rest of this page before you cite anything here.
What you are reading
The BERIL Research Observatory’s AI agents produced the 75 reports in this corpus. Most of them worked over the KBase Data Lakehouse: pangenomes, RB-TnSeq fitness assays, biochemistry, and environmental metadata. A few analysed data brought from elsewhere, user-provided sequencing among it, and each report states what it used. None of them generated new laboratory measurements. For each report the pipeline writes a summary, works out which synthesis pages it belongs on, and rewrites those pages to absorb it. A claim therefore gathers evidence from every project that supports it, rather than appearing once per project.
Every page states which kind it is, on the line above its title. Where a page cites evidence, that list names what it rests on and the numbered marks in the prose point into it. It is headed “Projects cited” where every citation is a project report, and “Sources cited” where a cross-project digest is among them.
Project reports. One page per research project, summarising it and linking to its raw report. These carry the evidence the rest of the site rests on: a citation points at one of these or at a digest below, and nowhere else.
Cross-project digests. Two pages — discoveries and pitfalls — reading across the whole corpus rather than summarising one project. Pages cite these the same way they cite a project report, and the citation says which it is.
Raw reports. The reports themselves, unedited, behind each summary.
Concepts. Recurring ideas, each gathering evidence from every project that speaks to it.
Entities. Specific named things: organisms, genes and pathways, compounds, methods, and datasets.
Topics. Hubs that group related concepts. Each opens with a literature-context section whose citations were checked against PubMed.
Conflicts. Places where projects in the corpus disagree, with the evidence on each side and the work that would settle it.
What the evidence label means
Synthesis pages carry a computed line under the title:
Evidence · 5 source projects · well corroborated · conflict on record · literature context
Every part of it is counted, not judged:
| Term | Means |
|---|---|
| N source projects | distinct projects the page’s prose actually cites |
| single-source | one project, so nothing here corroborates the claim |
| corroborated | 2 or 3 projects |
| well corroborated | 4 or more projects |
| conflict on record | the corpus records a disagreement spanning these same sources |
| literature context | the page opens with a literature-context section |
This is not a confidence rating and not a review status. “Well corroborated” means several projects here point the same way. It does not mean the finding is correct, and it does not mean anyone checked. Projects in this corpus share data sources, tooling and methods, so agreement between them proves less than it appears to. Several conflict pages exist because results that pointed the same way turned out to rest on one shared confound.
What is checked, and what is not
Every build checks these, and a failure stops publication:
- Every
[src:]citation resolves to a real source document. - Every number in a cited paragraph appears in a source that paragraph cites. The pipeline re-runs a page that fails against its own sources, and rejects the page if it fails twice.
- Every
wikilinkresolves. Dead ones become plain text.
Nothing checks these:
- Whether a finding is scientifically correct.
- Whether a statistical method suited its data.
- Whether the underlying research report reached a sound conclusion.
- Whether a synthesis fairly represents the projects it draws on.
One check runs when a page is written and not again at publication: the pipeline verifies literature-context citations against PubMed by PMID at that point. The literature context term only records that such a section exists.
The numeric check is real, and narrow. It proves a figure appears in a source the paragraph cites. It compares figures as text, so it misses a flipped sign or a changed unit, and it skips small integers. It never shows the source was right.
Known limitations
The pipeline merges two entity pages only when something identifies them as the same entity: a shared canonical name, a declared alias, or a shared external identifier. It never merges on similarity, which over these pages cannot tell two genes of one organism apart. Two pages naming the same thing without any of those signals stay separate. Entities that only one project cites do not appear here at all.
Reported budget figures undercount what a run actually spends, so the cost notes in the repository are lower bounds.
The data this rests on
Most projects in this corpus ran over the KBase Data Lakehouse, the data platform of the Department of Energy Systems Biology Knowledgebase (KBase). The pangenomes, fitness assays, genome annotations, amplicon surveys, metabolic models and metadata that those reports analyse are KBase’s, not this project’s. A few projects analysed data from elsewhere instead, and each one says so. Little on this site would exist without the lakehouse.
KBase asks to be cited as:
Arkin AP, Cottingham RW, Henry CS, et al. KBase: The United States Department of Energy Systems Biology Knowledgebase. Nature Biotechnology. 2018;36:566. doi:10.1038/nbt.4163
Cite that alongside anything you take from this wiki that rests on lakehouse data, which is all of it.
How to cite
Cite the specific page and the date you read it, since pages change when the corpus does.
BERIL Project. BERIL Knowledge Wiki: PAGE TITLE. Lawrence Berkeley National Laboratory. PAGE URL (accessed DATE).
For example:
BERIL Project. BERIL Knowledge Wiki: Fitness costs and conditional benefits of antimicrobial-resistance genes. Lawrence Berkeley National Laboratory. https://beril-doe.org/beril-wiki/concepts/antimicrobial-resistance-fitness-cost (accessed 2026-09-09).
For a finding, cite the underlying research report rather than the synthesis page that aggregates it. Every summary page links to its report.
License
This wiki is compiled from the BERIL Research Observatory’s research corpus, which is licensed AGPL-3.0, so the pipeline and the compiled pages carry the same licence. The licence grants you rights to the text. It makes no claim that the text is correct.
Source
The compiler, the editorial contract it enforces, and this corpus are at github.com/beril-doe/beril-wiki.