Methodology

The labels themselves — what S, H, A, M and R mean and why the scheme names the kind of study rather than ranking it — are on Evidence codes.

How a study gets in

Every study passes through the same four steps before it's added:

1. Source -- candidates are found via PubMed, prioritizing landmark discovery papers, systematic reviews and large human RCTs over secondary commentary; bioRxiv supplies preprints (always flagged as such, never treated as equivalent to peer-reviewed work), and ClinicalTrials.gov supplies ongoing human trials. Each new issue of bims-auttor -- the weekly "Autophagy and mTOR" reading list on biomed.news, selected by Viktor Korolchuk at Newcastle University -- is screened here as well: a reading queue, not an inclusion criterion, since an item still has to clear the same bar as anything else on this page.
2. Verify -- each citation's PMID, DOI, year and journal are confirmed against PubMed's own metadata before the entry is written; no citation is added from memory alone.
3. Label -- the study gets an evidence code naming the system the claim actually rests on (S/H/A/M/R), not a mark for how good it is. Until September 2026 these codes ran A–D; they were renamed because a lettered ladder is read as a school grade however it is captioned, and that quietly devalued molecular work that the pathway was built on. See the changelog.
4. Link -- the genes, drugs and outcomes the study mentions are connected to it in the graph, so the same paper surfaces wherever any of its subjects is explored.

Every DOI resolves to the publisher's own page (Nature, Cell, Science, NEJM, Lancet and others), so any claim traces back to the original paper in one click.

Inclusion & exclusion criteria

A candidate enters the corpus if it does at least one of these five things.

1. IN1 · Landmark mechanism -- establishes a mechanism in mTOR biology that the rest of the field builds on.
2. IN2 · Best available human evidence -- supplies the strongest human evidence that exists for a claim. The ladder runs systematic review (S), then randomised trial (H), then observational human study (H). A small trial qualifies when nothing stronger exists; its size and limits are then recorded in the study's own sample-size and limitations fields rather than left for the reader to guess at.
3. IN3 · Pathway coverage -- fills a pathway node that would otherwise have no supporting study, using representative animal (A) or mechanistic (M) work.
4. IN4 · Negative or null result -- reports a result that failed, or a relationship that was tested and not found. Included deliberately, since a database that only shows what worked misrepresents the actual state of the science.
5. IN5 · Synthesis anchor -- an authoritative review (R), used for orientation and framing. A review never stands as the only evidence behind a causal edge.

A candidate is left out if any one of these holds.

1. EX1 · Not a study -- secondary commentary, editorial or news piece rather than a primary study or review.
2. EX2 · Adds nothing -- duplicates a pathway node already covered by a stronger study and makes no new claim.
3. EX3 · Unverifiable metadata -- no persistent identifier that resolves to the source's own record: a PMID, a DOI, or an NCT number for a registered trial. PubMed is the first check, not the only one; papers awaiting indexing, preprints and registry entries are verified at their own source.
4. EX4 · Incidental to mTOR -- mTOR measured as one readout among many in a paper about something else.

Markers, not grades. Preprints (PP) and registered trials (RT) are admitted only as the best available evidence for a claim, and carry the PP or RT marker instead of an S/H/A/M/R code. A code names the kind of study; it is not a quality score.

What this doesn't guarantee

An honest limitation: the Atlas does not keep a screening log. Candidates that were considered and rejected leave no record, so no exclusion count can be quoted or inferred. That's a real weakness compared with a systematic review, where the screening flow is itself evidence that the search was unbiased -- here it isn't. Selection is a judgement call, made paper by paper, and it is not currently auditable from the outside. The Atlas should be read as a curated reading list with grades attached, not as a systematic review of the mTOR literature. Logging rejected candidates and the reason is a planned change.

Who reviews the selection: one person, the curator. There is no second reviewer or independent adjudication of borderline calls -- the usual safeguard against a single reader's blind spots is absent.

How often it updates

An automated job screens PubMed for new candidate mTOR studies and relevant conferences once daily, but the corpus changes only when a human accepts a candidate -- typically a handful of papers a month, sometimes none. Deliberate catch-up passes over older literature are the exception: they can add dozens of records at once, and each one is dated and described at /changelog/ (the largest so far, in October 2026, added 66 papers from 1997 to 2018). The exact snapshot of the live corpus (timestamp, dataset version and source commit) is published in the machine-readable metadata at /api/v1/meta.json, and every count on this site, including the ones above, is computed from that snapshot rather than typed in by hand. The date in each page footer is when that page was last rebuilt, so it can be earlier than the snapshot if the page's content did not change.

Corrections log

Every change to the method itself (how studies are selected, graded or read), with its date and the reason, is public at /changelog/.

← About the project · Corrections log · Data & citation