Guided biomedical discovery

MolecuRank

Move from literature to molecular candidates one reviewed decision at a time.

Research and education only. Results are evidence-ranked molecular hypotheses that require independent review and validation.
1
Define the evidence base

Search and review the literature

Why this step exists

This step defines which scientific papers are allowed to influence the rest of the workflow.

What MolecuRank does

MolecuRank inserts the disease phrase into your query, sends it to PubMed, and retrieves indexed titles, abstracts, journals, years, and PMIDs in relevance order.

What you should review

Check that the papers truly concern the disease, population, and mechanism you intended. Search matches are candidates for evidence—not automatically good evidence.

Explain the controls and terms
  • Disease or phenotype — The biological problem placed into the query.
  • Query template — The PubMed logic; {disease} is replaced by your disease phrase.
  • Human evidence filter — Adds the PubMed humans MeSH filter; it can exclude useful preclinical mechanism studies.
  • Maximum records — Caps how many PubMed results are retrieved for this run.
5 40
2
Adjust the analysis instruction

Extract and review disease targets

Why this step exists

This step converts unstructured papers into a reviewable list of proteins or genes that may matter in the disease.

What MolecuRank does

The configured MedGemma endpoint—or the clearly labelled fallback—looks for official target symbols, disease direction, supporting PMIDs, and evidence confidence.

What you should review

Confirm that each target is relevant, that “up” or “down” refers to the disease state, and that the cited papers support the statement. You can edit the table before continuing.

Explain the controls and terms
  • Extraction prompt — Tells the evidence analyst what relationships to find.
  • Maximum targets — Limits the breadth of the downstream chemistry search.
  • Disease direction — Whether the target appears increased, decreased, context-dependent, or unknown in disease.
  • Confidence — Strength of support in retrieved text; not proof that the target causes disease.
2 12
3
Set the chemistry funnel

Retrieve and filter compounds

Why this step exists

This step asks which known molecules have measured laboratory activity against the reviewed targets and removes weak or chemically unsuitable entries.

What MolecuRank does

MolecuRank maps targets to ChEMBL, retrieves measured activities, calculates RDKit properties from molecular structure, then applies your approval, potency, and QED thresholds.

What you should review

Inspect assay type, potency, target identity, molecule name, development phase, and chemistry. Target activity alone does not tell you whether the molecule changes disease biology in the desired direction.

Explain the controls and terms
  • Approved drugs only — Keeps molecules recorded as approved or with an approval year.
  • pChEMBL — Standardized target activity; higher usually means stronger potency when assays are comparable.
  • QED — 0–1 summary of drug-like chemical properties, not efficacy or safety.
  • Candidate limit — Maximum number carried into ranking after filters.
0 10
0 1
5 30
4
Make the scoring priorities explicit

Tune and run the final ranking

Why this step exists

This step makes the trade-offs explicit and orders the reviewed candidates under your chosen priorities.

What MolecuRank does

MolecuRank normalizes literature support, measured potency, QED chemistry, and clinical maturity; multiplies each by its weight; then applies an evidence-completeness adjustment.

What you should review

Look for candidates that remain strong under reasonable weight changes. A high score means comparatively well supported in this run—not likely to work in a patient.

Explain the controls and terms
  • Literature weight — Importance of disease-target support from Step 2.
  • Measured potency weight — Importance of ChEMBL target activity.
  • Chemistry weight — Importance of QED drug-like properties.
  • Clinical maturity weight — Importance of development phase or approval history.
  • Overall score — A within-run comparison from 0–100; not a probability or clinical effect size.
0 1
0 1
0 1
0 1