BioWeaver: detailed user guide
Scope: the shared-design release prepared on 2026-09-29. Start at BioWeaver. The Tools navigation connects every application; this guide is available from Documentation in that navigation.
The research-literature browser was added to the local build on 2026-10-05. Graph Explorer and its Python client were added to the local build on 2026-10-18. The bounded Drug Evidence Explorer pilot was added on 2026-10-24. See Graph Explorer for the shared browser/Python workflow. Study dossiers, molecular viewers and analysis importers were added to the local build on 2026-11-08. These additions have not been deployed by this work.
Contents
- Choose a tool
- Understand the evidence
- Prepare gene identifiers and lists
- Loom
- Evidence Map
- Unresolved
- Unseen
- Evidence Audit
- Orthology Workbench
- Evidence Trail
- Cross-species Comparator
- Evidence Changes
- Drug Evidence Explorer
- Review retired NCBI identifiers
- Projects, reports and downloads
- Move between tools
- Privacy and reproducibility
- Troubleshooting
- Study dossiers and molecular tools
- Analysis importers and examples
Choose a tool
| Tool | Start here when… | Input | Main output | Boundary |
|---|---|---|---|---|
| Loom | You need a gene, disease or model overview | Search term, symbol, stable identifier or browsing controls | Linked gene/disease pages and model evidence | Associations and orthology do not establish causality |
| Graph Explorer | You want to explore connected evidence from a gene list | Species, genes or a saved graph | Interactive network and portable evidence document | Graph proximity is not biological similarity; connections are source records |
| Evidence Map | You want disease coverage across species | Disease-name filter | Species-by-disease evidence grid | Curated and inherited coverage are different states |
| Unresolved | You want to examine clinical interpretation gaps | Search and constraint filter | Gene-level ClinVar/constraint summaries | It does not classify an individual variant |
| Unseen | You want to inspect gaps in reviewed human protein annotation | Gene/protein search and gap category | Reviewed UniProt annotation-gap rows | A missing field is not proof of biological absence |
| Evidence Audit | You have a human gene list | Pasted list or text/TSV file | Per-input resolution and source availability | Missing, ambiguous and unavailable remain distinct |
| Orthology Workbench | You need a list mapped between species | Source/target species and gene list | Candidates, supporting methods, exclusions and selections | It preserves source predictions; it is not a new orthology predictor |
| Evidence Trail | You need to inspect why a disease annotation exists | Species and one gene | Original relationships, donors and references | A donor annotation is not independent evidence in the recipient species |
| Cross-species Comparator | You want evidence for selected ortholog candidates together | Query gene, target species and candidate selections | Comparison of recorded evidence | More annotations do not establish functional equivalence |
| Evidence Changes | You need the status of historical comparison | Published baseline; locally verified comparison with gene/list, source and kind filters | Snapshot provenance; local JSON/TSV comparison reports | New comparison UI is not yet deployed; compare exact snapshot identities |
| Drug Evidence Explorer | You want measured compound and assay records for a selected target and its orthologs | BRAF, EGFR, PCSK9, TP53, BRCA1, PTEN, ARAF or RAF1 target; compound, assay, variant, year and endpoint filters | Protein-specific matrix, five count graphs, potency plot, records and exports | Bounded ChEMBL sample; orthology does not transfer activity or establish efficacy |
Evidence Map, Unresolved and Unseen use their Cloudflare Pages addresses because their advertised BioWeaver subdomains do not currently resolve. The shared menu uses the working destinations. Their data remains in separate static deployments.
Project/report and optional NCBI history controls are available in Audit, Orthology Workbench, Trail and Comparator. Do not expect those controls in every older tool or in Loom merely because all belong to BioWeaver.
Portal counts and source ages
Each count on a portal tool card is read from that tool's manifest when the site is
built. The drug card prints the anchor genes, ChEMBL targets and activity rows in
apps/drugs/data/manifest.json. A line under those counts names the anchor genes, the
ChEMBL release and licence from the same manifest, and states that activity is not
transferred across orthology. Evidence Map, Unresolved and Unseen read the totals in
their own data files. A card whose manifest is missing, or records no totals, prints
no count. It never shows a number from an earlier build.
The Source ages figure on the portal draws one bar per source in the data status: one source label at one version, shared by every tool that uses it. Each bar counts days from the oldest date any tool records for that source to the build date, which the scope sentence prints. A recorded release date is used first, drawn solid. A retrieval date is used otherwise, drawn outlined. A source with no usable date gets a hatched red bar labelled "no date recorded". Red marks only that error state. Download SVG saves the figure with the build date in its metadata. The data freshness page lists every version and date.
Research literature
Open Research literature in any tool's menu, or use the literature link on a Loom gene or disease page. Gene lookup accepts a symbol or stable identifier in the selected species. Disease lookup accepts a name or DOID; partial names offer exact terms to choose from. Disease results do not silently include descendants.
Source-cited references remain grouped with the disease, GO, phenotype or complementation annotation that supplied them. Negative and inferred evidence retain their labels and donor context. Show more source records exposes the remaining rows. Gene results count distinct PMID identifiers separately from reference identifiers, because multiple IDs can describe the same paper.
Broader PubMed discovery prepares external search links for a phrase, optionally including the selected species. Search matches are not source-curated evidence. Queries are sent to PubMed only when you follow a link. You can change the phrase here and refine it further on PubMed. BioWeaver does not fetch abstracts or full text in the background, infer paper titles, or promise exhaustive literature coverage.
Source counts and manifests describe the local release behind this browser. Allele and model disease evidence remains on Loom's annotation views. A gene with no matched source citations can still have a substantial literature.
Citations by source kind and ChEMBL drug documents
A gene lookup draws Citations by source kind above the results. Each bar is one kind (● disease, ◆ GO, ■ phenotype, ▲ fly complementation, ⬟ ChEMBL drug documents). Its label prints the number of source records and of distinct PubMed IDs for that gene, and the bar length is the record count. A PubMed ID cited by two kinds counts once in each. The scope sentence gives the totals across kinds. Every count is computed in the browser from the loaded records.
Select a bar to show only that kind in the result list. Select it again, or use
Show every kind, to clear it. Arrow keys move between the bars; Enter or Space
selects. The kind is kept in the link as #kind=drug, so a copied link reopens it.
Download SVG saves the figure with the gene, the kind and the ChEMBL release in
its metadata. A disease lookup has no figure, because its records arrive in pages.
The drug kind appears only for human HGNC genes. Its records are the ChEMBL
documents behind measured activities on that gene's own human target in the drug
evidence snapshot (BRAF: 65 documents with a PMID or DOI, ChEMBL 37). Documents
behind activities on a mouse or rat target in the same snapshot are left out, and a
mouse, fly or other non-human gene gets no drug kind at all. Activity is not
transferred across orthology. A human gene outside the eight drug anchors shows the
kind with zero records. Each drug card links to /drugs/#anchor=<HGNC id> and
carries the ChEMBL release and the CC BY-SA 3.0 licence link. Disease, GO and
phenotype citations appear as soon as the gene loads. Until the drug citations
arrive, the drug bar reads “loading”, and the drug cards then join the list. If the
drug citations cannot be loaded within 10 s, the drug bar is hatched and reads
“unavailable”, the note gives the reason, and no release or licence line is shown.
A #kind= link edited in the same tab applies in place. A kind the gene has no
records of is cleared.
Understand the evidence
Use Source publications in any tool's side menu to open its section of the resource bibliography. These papers describe upstream databases and methods. Retain the source versions in your export as well, and follow individual annotation references for the underlying experiments.
States are part of the result
| State or term | Meaning | What it does not establish |
|---|---|---|
| Recorded / present | A matched source record is available under that tool's rules | Biological truth, completeness or independence |
| No matched record | A usable source was checked but no record matched | That nobody studied the gene |
| Source unavailable | Required source material could not support that field | A negative result or zero |
| Ambiguous | More than one candidate remains | Permission to choose the first candidate |
| Excluded | A record/candidate was removed by stated selection rules | That the source never contained it |
| Uninterpretable | The available record cannot support the intended interpretation | A normal, low-risk or zero result |
| Zero | A recorded numerical value is zero | Missing or unavailable data |
| Curated | The annotation belongs to the curated category used by the source/exporter | That every displayed annotation represents a distinct experiment |
| Orthology-inferred | The association is inherited through an orthology relationship | Direct experimental evidence in the displayed species |
| Negative annotation | The source expresses a negated relationship | A positive association with a warning attached |
“Direct” can mean annotated to the exact disease term rather than a descendant. It does not necessarily mean experimentally curated rather than inferred. Treat term scope and inference as separate questions.
Phenotype descriptions may be source text. Do not assume they carry standardized ontology identifiers or support a similarity calculation. GO qualifiers, evidence codes and references matter: NOT and ND must not be interpreted as ordinary positive functional annotations.
Counts need a denominator
An input-row count, unique-gene count, annotation count and source-record count can all differ legitimately. A duplicated input remains two rows. One gene can have multiple proteins, transcripts or annotations. Multiple sources may repeat the same underlying evidence. Read the displayed scope and retain it in your report.
Source snapshots are dated datasets, not live queries against every upstream database. A current upstream page may legitimately differ from the saved release.
Prepare gene identifiers and lists
For Audit and Orthology Workbench:
- Paste one identifier per line, or open a text/TSV file.
- For a tab-separated table, choose the identifier column explicitly. The visible column control is one-based: column 1 is the first column.
- Enable the header option only if the first nonblank row is a header. The tool does not guess that a gene-like value is a heading.
- Choose species where the tool offers them. Audit is human-specific.
- Run the tool and inspect identifier resolution before interpreting evidence.
The parser supports quoted TSV cells, escaped quotes and newline-separated input. Blank rows are skipped. The selected identifier token is trimmed, while original row text is retained. An unclosed quoted cell is an input error. Comma-separated CSV is not the documented input format; convert it to TSV or paste a single column.
A deliberately illustrative input is:
BRCA1
BRCA1
__unmatched_public_example__
The duplicate tests preservation of repeated rows. The last token is an intentional non-gene example, not a biological identifier. Do not replace uncertain user input with a plausible symbol simply to obtain a resolved row.
For Trail and Comparator, start with one gene symbol or stable identifier and the appropriate species. Their single-gene input is not a list-upload substitute.
Loom
Entry: https://bioweaver.org/loom. Old loom.bioweaver.org paths redirect to
this location, preserving the path and query.
- Search a public example such as BRCA1 or browse diseases/models.
- Inspect the organism and stable identifier on a result before opening it.
- On a gene page, examine the available profile, orthologs, diseases and phenotypes.
- Follow a disease link to inspect the term-specific evidence and available scope.
- Use the relevant downloads when you need the displayed data outside the page.
- Preserve the URL and source context with your research notes.
Loom's existing local navigation includes Search, Diseases, Models, Advanced and Ask. These are sections of Loom, not separate BioWeaver tools. Ask is a separate query workflow; the static-tool claim that lists remain in the browser must not be generalized to text submitted to a server endpoint or model-backed query feature.
Stable IDs can appear in normalized URLs, such as /loom/gene/HGNC_1100. Use the
application's generated links rather than manually changing namespace punctuation.
Some source-only genes lack an indexed Loom page; the other tools deliberately
avoid inventing a working link for them.
Evidence Map
Entry: https://evidence-4y8.pages.dev/.
- Read the legend and introductory scope before reading the grid.
- Use “Find a disease” to narrow the disease rows.
- Compare the separate curated, inferred-only and missing states across species.
- Follow a disease link into Loom for more detail where available.
- Use “Show more” if the desired row is outside the currently rendered subset.
An apparently populated cell can represent inferred evidence. The grid is not a count of independent experimental confirmations, and its visual density must not be described as curated coverage without checking the legend.
Species stack
The Evidence states per species figure draws one bar per species. Each bar has one
cell per disease, so all bars have the same length. Each segment is the share of
diseases in one state: curated (●, solid), inferred from an ortholog (○, outlined),
no record (−, outlined grey) and not disease-curated in this release (×, hatched). The
count under each segment and the scope sentence above the bars are computed from
evidence-map.json when the page loads. They cover the whole map, before the disease
search.
- Select a species name to highlight its grid column. Every disease stays listed.
- Select a state segment to list only the diseases in that state for that species. The disease search then narrows within that list.
- Select the same item again, or Show all species and states, to clear the selection.
- The bars form one tab stop. Arrow keys move between them; Enter or Space selects.
- Download SVG saves the figure with its scope sentence and the current selection.
The page description and share preview print the same shares. The site build computes them from the payload, so they cannot drift from the data.
The Evidence Map has no drug column. A drug beside a disease row would read as a treatment claim, and ChEMBL target activity does not establish that a compound treats a disease. Drug evidence by gene is in the Drug Evidence Explorer.
Unresolved
Entry: https://unresolved-bqp.pages.dev/.
- Read the displayed ClinVar and constraint definitions and scope.
- Search for a gene of interest.
- Compare “All genes” with “Highly constrained only” when that restriction is relevant to the research question.
- Inspect the gene-level interpretation summary and linked gene context.
- Retain the source/version and denominator when reporting a fraction or count.
This tool examines human gene-level summaries. It does not determine whether an individual variant is pathogenic. Constraint is not an interpretation guarantee; the page's finding that constraint does not predict interpretability must remain part of the result, even when it weakens a proposed story.
LOEUF against uncertain share
The scatter plots one point per gene: gnomAD LOEUF on x and the share of submitted ClinVar alleles classified uncertain on y. Shape marks the constraint band: a triangle below the payload's highly constrained threshold (0.35), a diamond up to the constrained threshold (0.6) and a circle above it. Dashed lines mark both thresholds. The sentence above the chart counts the genes plotted, the genes left out for a missing LOEUF and the genes in each band, all from the loaded payload.
The chart is one tab stop. The arrow keys move between genes in LOEUF order, Page Up and Page Down move 50 genes, and Home and End jump to the ends. Enter or a click lists that gene's row, outlined and focused; a search or constraint filter that hides it is cleared first, and the status line says so. Download SVG saves the figure with its scope, thresholds and picked gene.
The page has no drug column. It is served from its own origin and cannot load the drug index from bioweaver.org. Drug evidence by gene is in the Drug Evidence Explorer.
Unseen
Entry: https://unseen-bpy.pages.dev/.
- Read the scope: reviewed human protein annotation and the fields inspected.
- Search the available gene/protein rows.
- Choose All gaps, Never observed, No function, No GO term, or No observation, function or disease. Disease is not one of the three gaps the combination chart counts.
- Interpret each gap in the context of its named UniProt field and source release.
- Follow a valid gene link when available and inspect the underlying source.
“Never observed” is the tool's source-defined protein-observation state; it is not a claim that no laboratory has ever looked at the gene. Missing function/GO text does not mean the protein has no function. The page's comparison showing little selection among its gaps is scientifically relevant and must not be omitted.
Gap-flag combinations
The chart counts each protein in exactly one combination of its three gaps: never observed as a protein, no FUNCTION comment, and no GO term. A missing disease is not a gap here, because it never puts a protein on the page by itself. Bars are sorted by count, largest first. Under each bar a dot matrix names the combination: a filled dot for a gap it has, a hollow ring for one it lacks, with the filled dots joined by a line. Each row label carries that gap's total. The sentence above the chart states how many proteins were counted, how many combinations occur and each gap's total, all computed from the loaded payload.
The chart is one tab stop. The arrow keys move between combinations, Home and End jump to the ends, and Enter, Space or a click lists only the proteins with exactly that combination. The pressed column is outlined, its count underlined, and it carries aria-pressed. Picking it again, or Show every combination, clears it. A combination resets the gap chips to All gaps, and a chip clears the combination; the search box narrows either. Download SVG saves the figure with every combination count and the current selection.
The page has no drug column. Its rows are UniProt accessions, some without an HGNC id, so drug states would first need an accession sweep that maps every row to a gene. The page is also served from its own origin and cannot load the drug index from bioweaver.org. Drug evidence by gene is in the Drug Evidence Explorer.
Evidence Audit
Entry: https://bioweaver.org/audit/.
- Enter a human gene list or open a text/TSV file.
- Set column/header controls and select “Audit evidence”.
- Inspect resolved, unresolved and ambiguous inputs first.
- Choose an evidence category: identifier resolution, constraint, ClinVar, proteins, diseases, mouse models or drug evidence (ChEMBL).
- Choose a state filter if you want to isolate missing records, unavailable sources, ambiguous matches, excluded records or uninterpretable results.
- Expand available details to identify which source supports a cell.
- Download visible rows as TSV for the filtered table, or a session manifest for input/options provenance. Save a project for the archived evidence and notes.
The visible filter changes the view. A project retains its captured evidence and filter settings; it is not interchangeable with a visible-rows TSV. Do not assume that a short filtered export contains every input or all underlying detail.
Mouse-model summaries require their stated production/allele evidence. A missing particular IMPC state is not proof that no knockout exists in MGI or elsewhere.
Drug column, state bars and LOEUF histogram
The matrix has a sixth column, Drug evidence, read from the /drugs/
ChEMBL index (/drugs/data/manifest.json, index schemaVersion 2). Each input
row gets one state:
| Glyph | State | Meaning |
|---|---|---|
| ● | ChEMBL activity records | An anchor gene with collected activities. The cell links to /drugs/#anchor=<HGNC id>. |
| ◐ | ChEMBL target, activity not collected | A human single-protein target exists. Mechanism count and highest phase are shown; absence of activity rows says nothing. |
| ○ | Not in ChEMBL target layer | No exact human single-protein target in the snapshot. That is not "no drugs". |
| — | Not evaluated | An unresolved input, or any identifier that is not a human HGNC id. Activity is not transferred across orthology, so a mouse or fly row never inherits its human ortholog's state. |
| ! | Drug index unavailable | The index failed to load, or did not load within 10 s. Every drug cell shows this; the other five columns still work. |
The ladder ■■■■ shows the highest mechanism phase from 0 to 4 (◧□□□ is early phase 1). The note under the matrix names the ChEMBL release, links CC BY-SA 3.0 and states that activity is not transferred across orthology.
The first table row holds a state bar for each of the six columns. A segment shows a state glyph and the number of input rows in that state, counted over the rows that match the search and LOEUF filters; a repeated identifier counts once per row. The bars ignore the source and state filter they drive, so every segment stays selectable. Selecting a segment (click, Enter or Space) sets "Filter source" and "State" to that column and state; selecting the pressed segment again clears the state. "Download state bars SVG" saves the counts.
The LOEUF histogram bins the recorded gnomAD LOEUF of the current rows in steps of 0.2 up to 2.0, then one open bin. Rows without a value, with an uninterpretable constraint record or with an unresolved identifier are listed under "Not binned", never as zero. Selecting a bin filters the matrix to rows in that range; selecting it again, or "Clear LOEUF filter", removes it. The histogram follows the source, state and search filters but not its own range. The LOEUF filter is recorded in the session manifest and project view; it is not written to the URL. Opening a project does not restore it, and a new gene list, a new run or "Clear view filters" clears it. Keyboard focus stays on the segment or bin just selected.
Matrix TSV rows for the drug source carry drug_mechanisms, drug_max_phase
and source_release (the ChEMBL release). drug_mechanisms is filled for
ChEMBL activity records, and for an unqueried target only when the index lists a
nonzero count; otherwise it is blank, never 0. Matrix JSON, figure metadata and
the session manifest carry a drugIndex block with the release, licence,
licence URL, share-alike note and orthology sentence. The visible-rows TSV adds
drug_state, drug_mechanisms, drug_max_phase, chembl_release and
drug_licence; the last two are filled on every row.
Orthology Workbench
Entry: https://bioweaver.org/orthology/.
- Choose From species and To species.
- Enter the list and column/header settings.
- Select “Map gene list”.
- Inspect method support and the available reciprocal/method filters.
- Review excluded candidates as well as included candidates when explaining a selection. A candidate excluded by your settings is still a source record.
- Select candidate models explicitly. Keep one-to-many mappings visible.
- Choose the export scope: all candidates and exclusions, included candidates, or your selections.
- Download mapping TSV, unique targets, a session manifest, or a project/report.
“Unique targets” intentionally deduplicates a list for downstream use; it is not
a substitute for the mapping table when input duplicates or many-to-one mappings
matter. Directional flags such as Yes_Adjusted retain their original meaning.
Changing a query can clear selections; save the project before starting another.
This is a workbench over the supplied Alliance/DIOPT-derived predictions. It does not calculate new sequence alignments or claim better orthology prediction accuracy.
Drug chips, support histogram and retention curve
Each candidate row carries two ChEMBL chips. The chip under the input gene
describes the query; the chip under the target gene describes the candidate.
Each reads its own gene: only a human HGNC id is looked up in the /drugs/
index, so for a human-to-mouse mapping the BRAF query chip reads "ChEMBL activity
records" and links to /drugs/#anchor=HGNC:1097, while the Braf candidate chip
reads "Not evaluated". For mouse to human the roles swap. Activity is not
transferred across orthology. A glyph (●, ◐, ○, —, !) and the state name carry
the state besides colour. The line under the table names the ChEMBL release and
links the CC BY-SA 3.0 licence. If the index fails or takes more than 10 s, human
chips read "Drug index unavailable" and mapping, filters and exports still work.
"Candidates by number of supporting sources" counts every candidate in the table, included or excluded, by its support. Bins at or above the minimum support are solid and labelled "retained"; bins below are outlines labelled "below threshold". Moving the slider redraws them. Selecting a bin (click, or Enter on the focused bin; the bins share one tab stop and move with the arrow keys) sets the minimum support to that value and moves the slider.
"Candidates retained by minimum support" plots how many candidates survive each threshold from 0 to the largest support plus one, holding the reciprocal and required-source filters as set. A dashed rule and a diamond labelled "≥ N: count" mark the current threshold; at that threshold the curve equals the number of included candidates. Enter on a curve point sets the threshold too. Both figures print a scope sentence counted from the loaded data and download as SVG with the current minimum support in their metadata.
Evidence Trail
Entry: https://bioweaver.org/trail/.
- Choose a species and enter one gene.
- Select “Trace evidence”.
- Inspect the annotation-type filter. Its initial value is Orthology-inferred; choose All annotations if you want the complete set of categories.
- Filter by disease name or DOID where useful.
- Read the relationship, polarity, donor genes and references for each record.
- Distinguish the exact-term local-curation check from the original inferred record.
- Download filtered TSV, evidence JSON or a session manifest; save a project to retain the captured evidence and your interpretation separately.
Donors come from the source's basedOnGenes data. The donor species, disease term
and reference are part of the explanation. A positive-looking relationship name
alone is insufficient to determine polarity: source-generated relationship text
can carry negation. The tool retains that distinction.
Provenance stack, donor species and the pharmacology hub
"Disease records by provenance" splits every disease annotation record of the
gene into Curated (●), Orthology-inferred (◇), Negative (−) and Unstated (?). It
counts all records, before the sidebar filters, so the bar does not shrink when
you filter. Negative means the Alliance's generatedRelationString is negated,
for example is_not_implicated_in or does_not_model. Each segment shows its
glyph and count; the legend lists all four states, zeros included. Selecting a
segment (click, or Enter on the focused segment) sets the annotation-type filter
to that state; selecting it again shows all annotations. For human BRAF the
snapshot holds 46 records: 32 curated and 14 orthology-inferred.
"Donor species of orthology-inferred records" counts each inferred record once
per species among the donor genes it names. A record naming donors in two
species counts in both, so the bars can sum above the number of inferred
records; the scope sentence says how many do. Records that name no donor are a
hatched band labelled "No donor gene named", never a zero bar. Enter on a species
bar opens the source browser on that species' donor genes (search
taxon 10090). Both figures print a scope sentence counted from the loaded data
and download as SVG.
For a human HGNC gene with ChEMBL activity records or mechanism rows, the
provenance graph adds a "Pharmacology" node joined to the query gene and to
nothing else. Drug evidence about a protein does not imply treatment of any
disease the gene is annotated to, so no line joins the hub to an annotation.
BRAF's hub reads 14 mechanisms and 9,834 activities; Enter on it shows the drug
block with a link to /drugs/#anchor=HGNC:1097. A gene of any other species,
mouse Braf included, has no hub: activity is not transferred across orthology.
The note above the source browser names the ChEMBL release and links the CC
BY-SA 3.0 licence. If the drug index fails or takes more than 10 s, the note
reads "Drug index unavailable" and annotations, figures and exports still work.
Cross-species Comparator
Entry: https://bioweaver.org/compare/.
- Choose the query species and enter one gene.
- Choose target species and select “Find ortholog candidates”.
- Inspect candidate identities and supporting orthology evidence.
- Select the candidates you actually want to compare.
- Select “Compare selected candidates”.
- Inspect each evidence category and its missing/ambiguous/quality state.
- Download evidence TSV, comparison JSON or a session manifest, or save a project.
GO annotation totals are not functional-distance scores. Computational and experimental evidence are different categories. NOT and ND have special meaning. Human gnomAD constraint versions must retain their version/transcript/quality context. FlyBase complementation retains its direction, assay and references; it does not prove full functional equivalence between two genes.
Empty evidence for a selected candidate remains an explicit result. It must not silently remove the candidate from the comparison or make another one look stronger.
Drug blocks and paired evidence bars
Each comparison column has a "Drug evidence (ChEMBL)" block. A human HGNC gene is
looked up in the /drugs/ index and shows its state, mechanism record count,
highest mechanism phase as a four-step glyph (■■■■ is phase 4), the ChEMBL
activity total, and, for a gene with collected records, a link such as
/drugs/#anchor=HGNC:1097. A gene from any other species reads "Not evaluated":
activity is not transferred across orthology, so mouse Braf never inherits human
BRAF's records. Each block carries the ChEMBL release, the CC BY-SA 3.0 link and
that sentence.
"Paired evidence bars" sets the compared genes side by side on three drug measures (mechanism records, max phase, ChEMBL activities) and five column counts (curated and inferred disease terms, positive and experimental GO terms, phenotype text pairs). Each measure has its own scale; max phase is scaled to 4. A gene with no value for a drug measure gets a full-width hatched band labelled with the reason, never a zero bar. The bars share one tab stop: arrow keys move between them and Enter moves focus to that gene's column. "Download SVG" saves the figure with its scope and the drug index metadata.
The evidence TSV adds drug_state, drug_mechanisms, drug_max_phase,
chembl_release and drug_licence to every row. The comparison JSON adds
drugIndex (release, licence, orthology note) and one drugEvidence row per
gene; the session manifest adds drugIndex. If the drug index fails or takes
more than 10 s, the blocks read "Drug index unavailable" and the rest of the
comparison and its exports still work.
Evidence Changes
Entry: https://bioweaver.org/changes/.
Implementation status (2026-10-15): the published release remains baseline-only. The comparison interface described below is implemented and tested locally, but is not yet deployed. A real pair has now passed independent reconciliation and raw-source checks; see the refresh record.
The interface compares exported disease records under policy disease-v1.
These records include source relationships, polarity, donors and references, plus
derived donor-indexing and exact-term curation fields. A difference in an exported
record is not necessarily a new upstream assertion. GO, orthology and numeric
constraint changes are outside this comparison.
The validated local pair re-exports raw Alliance 9.1 and 9.2 with the same exporter and current index/catalogue. Its normalized 9.1 “before” snapshot is a separate artifact from the preserved, originally published baseline. Compare the named snapshot identities and provenance; the comparison must not be described as a byte-for-byte history of what the old website displayed.
When a verified pair is available:
- Read the comparison state and both snapshot identities before interpreting events.
- Filter by gene symbol or stable entity ID, or paste a local list separated by spaces, commas or newlines. Symbols match literally, without case sensitivity, across organisms. Use stable IDs to distinguish genes; aliases and retired IDs are not resolved here. The list stays in your browser.
- Narrow by source and change kind. Source-wide availability and comparability issues remain visible when filtering gene IDs. No matching event does not establish that a gene's biology is unchanged.
- Browse the paginated results and expand “Inspect before and after evidence” to read every exported field. The current Trail exporter does not retain stable annotation IDs, so differences appear as record additions/removals rather than guessed before/after pairs. A “Record fields changed” event requires a retained, shared source annotation identity.
- Download a filtered JSON report or TSV. Both retain the snapshot identities, policy, source fingerprints, filters and complete matching records. Exports include every matching event, not only the visible page. JSON also preserves full snapshot and exporter metadata once per report, including when no events match. TSV uses compact source fingerprint and release columns instead of repeating the full metadata on each row; it is disabled for an empty result.
- Use “Restore filters from a JSON report” to reopen a saved selection for the same snapshot pair and policy. This restores filters against the verified hosted comparison; it does not replace evidence with records from the file. This is a separate format from the project bundles in Audit, Workbench, Trail and Comparator.
Removal from these snapshots is not retraction. An unavailable source produces an availability/comparability explanation, not biological removal counts. Changed exporters or configurations also make the affected comparison partial. The build checks the current Trail identity, archived dependencies and recorded provenance, and separately verifies event completeness against the archived exported records before accepting a comparison. That check does not establish that upstream extraction was complete.
If only a baseline exists, the page offers its provenance download and a link to Evidence Trail; comparison controls stay hidden. Preserve reviewed evidence in a Trail project while a verified historical comparison is unavailable. No recurring watchlist, notifications or automatic monitoring are provided by this interface.
Organisms and change states
Two figures above the gene chart count record events in the loaded comparison. Both are computed in the browser from the comparison file on every filter change.
- Record events per organism. One row per organism, keyed on the record's
subjectTaxon, with a stacked bar split into added (solid,+), removed (hatched,−) and field changes (outlined,Δ). All rows share one scale. The count under each bar is printed with its glyph, so the figure reads without colour. Counts follow the gene, list and source filters, before the organism and state selection. - Record events by change state. One stacked bar of the three states for the current filters, including the organism selection.
Select an organism name, a segment, a count or a state-bar segment to filter the
gene chart and the record list. Select it again, or use “Show all organisms and
states”, to clear it. Arrow keys move between the figure's buttons; Enter or Space
selects. The selection is kept in the link as #organism=7227&kind=removed, so a
copied link reopens it, and it is saved in the filtered JSON report. The change kind
selection is the same value as the sidebar's “Change kind”. Each figure has a
“Download SVG” button whose file names the snapshots and the selection.
Source-level availability and comparability events have no organism or record state and are left out of both figures; the scope sentence says how many.
There is no drug surface on this page. BioWeaver holds one ChEMBL release, so no drug evidence change exists to compare.
Review a research decision in Evidence Trail
The decision-review workflow is prepared for preview evaluation and has not yet been released on the production site.
To practice before creating your own project, expand Try a worked example of decision review in Evidence Trail. Download the example and open it using Open project. It contains real archived annotations with illustrative reasoning, and no saved researcher review. Check the hosted comparison, inspect the differences, record your own assessment and download the result. Save an existing project before opening the example. The example is generated only when a suitable compatible historical pair is available.
- Run Evidence Trail or open a saved Trail project in “Project and report”.
- Under “Research decisions and evidence review”, select “Choose supporting records”. The selection includes the archived annotations hidden by display filters. Select the exact records you used; negative and inferred annotations remain labelled and can be part of your reasoning.
- Describe your decision and explain why the selected evidence matters, including assumptions and limitations. Select “Save decision in this project”, then Save project to download it. Work in a tab is not persistent storage.
- Select “Check decisions against hosted comparison”. This downloads public comparison data; your project, notes and decisions remain in the browser.
- Inspect linked removals or changes and separately listed additions for the same gene and exact disease term. An addition is not automatically supporting evidence. Partial removal of duplicate records reports a count change because the selected occurrence cannot be identified. A missing record does not establish retraction.
- Record whether you retain, revise or defer the decision, with a rationale. Revising also requires the revised decision text. An unassessable comparison only permits deferral. Download the updated project or readable HTML report to retain the review.
The decision must use the comparison's exact before snapshot, with matching source, exporter and configuration. A project from the after snapshot waits for a later compatible pair. An old published snapshot is not interchangeable with the normalized 9.1 snapshot, even when both use the same release. Older project formats remain readable, but missing configuration provenance can prevent comparison. No path through intermediate snapshots is inferred. Reports label unassessable states explicitly.
Reviews append to the original decision and keep its original baseline. A revised sentence does not automatically replace its supporting evidence. Start a new project with a current analysis when you want to adopt new evidence as a baseline. Existing decision projects remain separate when you change analysis inputs or run another query; download them before using “Start a new decision project”. Both project import and export are limited to 25 MiB. Imported files are user-controlled local records, not signed or independently authenticated review attestations.
Drug Evidence Explorer
Entry: https://bioweaver.org/drugs/.
Choose one of the eight selected human targets, then narrow the sampled ChEMBL activities by target protein, assay format, measurement type, assay species, variant annotation, publication year or compound/publication search. The compound-by-protein matrix counts records for the exact assayed protein; it does not copy activity across an orthology relationship. Its bounded, two-direction scroll area keeps column headers and compound labels visible.
The five interactive count graphs cover measurement type, assay species, variant annotation coverage, publication year and separate mechanism actions. Select a measurement bar to filter activity records; a mechanism-action bar filters the separate mechanism table and JSON export. The potency plot includes only exact, positive nM IC50, EC50, Ki and Kd values that pass the displayed source checks; points remain grouped by protein, assay species, variant, assay format and endpoint. Selecting a potency point scrolls to and frames its matching activity row.
This is a bounded sample, not a complete literature search or efficacy ranking. Read the per-target query-window counts and truncation before treating an empty slice as meaningful. See the pilot measurements and refresh guide. For step-by-step use and definitions of every graph, see the detailed workflow.
Review retired NCBI identifiers
Available in Audit, Orthology Workbench, Trail and Comparator.
- Enter explicit identifiers beginning
NCBIGene:orNCBI_Gene:and select the relevant species. Symbols and bare numbers are outside this history workflow. - Select “Review retired NCBI IDs”.
- Read the result and complete replacement chain.
- If a unique available terminal replacement is offered, choose “Accept replacement” or “Keep unresolved”.
- Run the tool again to apply the decision. The original input text remains intact.
- Save the project to retain the decision, chain and history snapshot reference.
| History result | Interpretation and next action |
|---|---|
| Current | Already owned by the current catalogue; normal resolution takes precedence |
| Replacement | A terminal replacement has one current owner; explicit review is possible |
| Withdrawn | The chain terminates without a replacement; do not invent one |
| Ambiguous | Conflicting replacements or multiple current owners; no automatic winner |
| Cycle | History loops; no safe terminal replacement |
| Replacement unavailable | A replacement is named but not uniquely available in the required catalogue |
| Taxon mismatch | The species context does not match the history rows |
| Not found | No applicable history record; this is not proof the identifier never existed |
Accepting a suggestion does not override an already resolved current stable ID. Opening a saved project clears active history suggestions from the previous work. A fresh analysis requires renewed review; the archived report remains readable.
The retained human-history input contains 166,633 retired IDs, of which 21,221 terminal replacements have a unique owner in the measured catalogue. These are snapshot-level measurements, not a promised recovery rate for your list. Exact denominators and chain outcomes are in the measurement report.
Resolution timeline
After a review, the panel draws a timeline above the identifier rows.
- The x axis runs in years across the NCBI discontinue dates of the reviewed
identifiers. Each date comes from the
Discontinue_Datecolumn ofgene_history. - Each retirement step is one mark. A chain of two retirements draws two marks joined by a dashed line.
- Each identifier sits in the lane of its final outcome. Each lane has its own glyph and mark shape: ● replaced by a catalogue gene, ◆ replaced but the replacement is not in the catalogue, ■ withdrawn, ▲ ambiguous. Cycle (↻) and taxon mismatch (≠) lanes appear only when an identifier lands in them.
- Each lane label prints its identifier count. The scope sentence above the chart states how many reviewed identifiers are retired, how many steps are drawn and the history snapshot. Current and not-found identifiers have no retirement date; they are counted in the sentence and listed in the rows only.
- NCBI records no merge flag. A merge and a replacement are both a row with a replacement id, so both count as replacements.
- Select a mark, by click or with Enter or Space, to move focus to its identifier row. The marks share one tab stop; the arrow keys, Home and End move between them.
- “Download SVG” saves the timeline with the snapshot id in its metadata.
- If no reviewed identifier has a usable date, the panel draws one bar per outcome instead and says how many retired identifiers lack a date.
The rows name each outcome in the same words as the lanes.
Projects, reports and downloads
Save a project
- Run the tool and complete the selections you want to retain.
- Enter the research question and researcher notes in “Project and report”.
- Select “Save project”. Keep the JSON file in storage appropriate for its contents.
- Select “Download report” for a standalone readable HTML report.
- Use “Print report” if you want the browser's print workflow. Actual print-dialog behavior has not yet been verified in the latest release environment.
Notes are researcher-authored interpretation. They are not presented as source annotations. The project includes original inputs, row identities, decisions, evidence, view settings and a closed set of source-snapshot dependencies.
Reopen a project
- Open the same tool that created it.
- Use “Open project” to select the JSON file. The current import ceiling is 25 MiB.
- Read the restored archived report and restored input/control settings.
- Edit the notes/question and save again if only those need changing.
- Run the tool again only when you intend to analyze against current hosted data.
Importing a project does not fetch evidence from URLs embedded in the project. A project from a different tool is rejected with the corresponding tool context. Unsupported schemas, invalid dependencies and malformed content are rejected. The size ceiling is a parsing bound, not a guarantee of fast printing at that size.
Choose the right artifact
| Artifact | Best use | Important limit |
|---|---|---|
| Project JSON | Preserve evidence, original rows, decisions, notes and context | Reopen in its originating tool; not a universal cross-tool interchange format |
| HTML report | Read, share or print captured evidence and interpretation | Not an executable fresh analysis |
| TSV | Spreadsheet work with the stated export scope | Flattened output; inspect filters and export mode |
| Evidence/comparison JSON | Preserve structured output for that tool | Not necessarily the project schema or editable notes workflow |
| Session manifest | Identify input/options/source context | Does not by itself guarantee all old evidence remains hosted |
| Snapshot archive | Operator retention of verified public datasets | Preserves derived results; raw downloads and historical code need separate retention |
Saved HTML reports carry their content and styling. Live URLs in them still need network access when followed. Saving a report does not archive every linked paper or database page.
Move between tools
Every tool has the same BioWeaver navigation. On a wide screen it stays in the left sidebar; on a smaller screen open Tools near the top of the page. The current tool has a visible marker. Escape closes the compact navigation and returns keyboard focus to Tools. All tools, Documentation and About & sources are available below the tool groups.
Appearance offers System, Light and Dark. System follows the operating system; the other options override it. Preference storage is local to each origin, so a separate Pages tool may have a different saved preference. If browser storage is blocked, the control still works for the current page. Without JavaScript, the expanded native navigation remains usable.
A sidebar link is navigation. It must not silently transfer your list, notes, accepted replacements or private project to another tool. Save work before leaving an analysis you need. A browser back button is not a durable storage mechanism.
Useful manual workflows include:
- Audit → Trail: identify a gene whose disease evidence needs explanation, then inspect that gene and the relevant annotation category in Trail.
- Orthology → Comparator: identify candidate orthologs, then explicitly select the candidates in Comparator and review their recorded evidence.
- Evidence Map → Loom: identify a disease/species coverage state, then inspect the relevant disease page and distinguish exact-term/inherited evidence.
- Unresolved or Unseen → Audit: take a public candidate into a broader source review, retaining the original gap definition rather than treating it as a score.
These workflows do not imply that every step currently has an automated handoff. Project files from one tool should not be imported into a different tool as though their evidence and selections have the same meaning.
Privacy and reproducibility
Static research-tool lists are processed in the browser; public catalogues and partitions are downloaded as needed. Project files and notes are saved locally. This is not a claim that the browser makes no network requests, or that every BioWeaver application has identical request behavior: Loom uses server endpoints.
The site does not add analytics or require an account. Downloaded reports can contain the identifiers and notes you entered. Sharing a report shares that content; inspect it before sending it outside your intended audience.
For reproducible work, retain the project, source snapshot IDs, source releases, view/filter settings, original input, accepted history decisions and your notes. An operator can additionally preserve the complete verified snapshots. A later re-analysis can differ because data, catalogue ownership or exporter behavior changed; it should not overwrite what the old report said.
Troubleshooting
| Symptom | What to check | Interpretation |
|---|---|---|
| Everything is unresolved | Species, identifier namespace, selected column and header setting | Input mismatch is not evidence absence |
| One gene is ambiguous | Candidate IDs and aliases in the selected species | Use an appropriate stable ID after review; do not select an arbitrary owner |
| No visible rows after a run | Active source/state/disease filters | The filter may hide rows without deleting them |
| An evidence category is unavailable | Source/snapshot section and download status | Do not replace it with zero |
| An NCBI replacement cannot be accepted | Terminal owner, selected taxon and history status | A named replacement may be absent or ambiguous in the current catalogue |
| A download fails checksum validation | Retry after reload; report the named artifact/snapshot if persistent | Mixed or corrupt bytes are refused rather than trusted |
| Snapshot metadata changed while saving | Preserve the tab, then review/re-run against consistent metadata | The project cannot honestly mix incompatible dependencies |
| A project will not open | Originating tool, file size, schema and complete JSON | A TSV or session manifest is not necessarily a project |
| Controls reset after changing the query | Whether the old result belongs to the changed input | Old evidence must not remain attached to new input |
| Print is incomplete or awkward | Use the downloaded HTML report; retain project JSON | Current automated checks do not certify print layout |
| A legacy subdomain fails to load | Portal link, current runbook and deployment/DNS status | A website failure says nothing about biological evidence |
| Loom fails while a static tool works | Operator read-budget/worker checks | Static tools and Loom have different serving dependencies |
When reporting an issue, include the tool URL, public example input if possible, selected species/filters, expected and actual behavior, error text and snapshot ID. Remove private notes and identifiers before posting a public issue. Operators should reproduce on local data before spending production database reads.
Graph Explorer
Choose the species for your starting identifiers, paste one gene per line and select Explore evidence. Repeated inputs remain separate in the saved input decisions; ambiguous identifiers are not silently assigned. Inspect those decisions below the graph.
Circles represent genes, diamonds annotation occurrences, rounded boxes diseases and outlined boxes references. Solid, dashed and dotted connections distinguish curated, inferred and negative annotation states. Disease associations are not causal pathways. Use the annotation list or inspection menu for keyboard access to records, including orthology method support and best-match flags.
Select a loaded gene, choose a target species and select Add ortholog candidates. The workspace adds that snapshot's directed candidates and their disease records. Filter by annotation type or search disease/gene labels; references and donor context can be shown separately. Zoom or focus a selected node to inspect a dense region. A page limit keeps the drawing manageable and does not remove records from saved data.
Save complete graph retains all inputs, loaded evidence and source manifests. Download filtered records includes matching annotations across all pages. PNG exports show the current bounded network and name the snapshots. Open a saved graph to restore its evidence without fetching current data; expanding it requires matching source manifests.
The Python client and format guide explains how to install the free client, read local or hosted snapshots, generate a graph and open it in the browser. Browser-saved files can also be read and validated in Python. Saved graphs are editable evidence documents, not signed source attestations.
Inspecting orthology method support
Open Orthology Workbench → Explore candidate-by-method support. Choose one gene and species pair, then select either the Alliance stringent subset or the explicitly versioned prepared DIOPT collection. The heatmap shows which methods or curator sources report each candidate. A dot means support was not reported, not a negative result. The visible columns cover methods found among the returned candidates; each row retains its original score denominator.
Select a candidate name to inspect stable identifiers and forward/reverse best-match flags. Ties and alternative candidates are retained. Requiring a method or raising the minimum support filters candidates and reports how many were excluded. Switching providers clears the previous view; the two providers' overlapping evidence is never added together. Genes outside the prepared DIOPT collection have an explicit not prepared message.
JSON downloads contain source receipts, filters and records. TSV downloads begin with
a # metadata: JSON comment followed by the data table; configure your importer to
skip comment lines. The metadata remains present for empty filtered results. SVG
figures show the displayed candidate page, while record exports cover all filtered
candidates. Protein and domain alignments remain available through DIOPT.
Exploring PANTHER family trees
Open PANTHER Tree Explorer from the Tools menu. Choose one of the selected
families and search for a gene symbol, source identifier such as HGNC:11179, or
source organism code. Search opens matching paths and underlines matching tips.
A missing result describes this selected family only.
Family choices show example indexed genes. For TP53, choose PTHR11447 and search
HGNC:11998; related TP63 and TP73 remain separate tips with their own identifiers.
For PTEN, choose PTHR12305 and search HGNC:9588; for BRCA1, choose PTHR13763
and search HGNC:1100; for BRAF, choose PTHR44329 and search HGNC:1097.
The BRAF-containing family is large, so search is a useful starting point. Its members
are not all asserted to be BRAF orthologs.
Select a node to inspect annotations and original identifiers. Use Expand this clade, Collapse this clade, Overview, or Expand all to change the view. Triangles mark collapsed clades. Branch spacing switches between equal topology steps and source lengths; the latter are not years or confidence. Zero-length branches can overlap; topology spacing separates successive nodes.
Download the original tree, complete family JSON with provenance and current view state, or an SVG of the visible tree. Collapsed descendants remain in the downloads. Source organism labels and unresolved identifiers remain visible even when Loom has no matching gene page.
Each tip carries a ChEMBL drug glyph beside its label: ● activity records, ◐ ChEMBL target with activity not collected, ○ not in the ChEMBL target layer, — not evaluated. Only tips resolved to a human HGNC gene are looked up; a mouse or fly tip reads "Not evaluated" even when the human gene beside it has records, because activity is not transferred across orthology. A ● glyph links to that gene in the Drug Evidence Explorer. The Drug evidence on tips panel under the tree counts the tips per state for the open family; select a bar, or focus it and press Enter, to ring those tips and open their clades, and select it again to clear. The panel names the ChEMBL release and links the CC BY-SA 3.0 licence. If the drug index does not load, the tree still works and human tips read "Drug index unavailable".
Exploring fungal mappings
Open Fungal Mapping Explorer from the Tools menu. The PomBase view starts with fission-yeast cdc2 and shows its separate human and budding-yeast mappings. Search either partner's name or identifier and select a connection to inspect the original source row. A Loom link appears only for a uniquely resolved indexed target. PomBase genes link to their own source pages for current curation and references.
Switch to CGD · CGOB groups to inspect fungal membership. Numbered matrix columns refer to the species/strain key beneath the graph. Select a row to see all identifiers, including unmatched legacy identifiers. A dot means no member is recorded in that cell; it is not evidence of gene loss. The source file date is visible above the graph.
The Species presence histogram counts, for the current search, how many species each record is present in. CGOB bins count group rows by species with a recorded member; the two Candida albicans strain columns count as one species, so 15 columns give bins 0 to 14. PomBase bins count fission-yeast genes: fission yeast itself plus human and budding yeast when a curated mapping reaches them, so bins run 0 to 3. The mapping target filter narrows the gene set but not a gene's count. Select a bar, or press Enter on it, to show only those records; the selected bar is marked ▶. Select it again or press Show every presence count to clear. Download presence SVG saves the histogram with its scope sentence and source metadata. Fungal genes have no ChEMBL human target, so this explorer shows no drug data.
Both graphs support keyboard selection with Tab and Enter, paging, and horizontal
scrolling on small screens. JSON and TSV export every matching record with source
receipts and the current filters. TSV begins with a # metadata: comment; configure
your importer to skip comment lines. SVG exports the displayed graph page, including
its strain key and provenance. Source row numbers identify the downloaded file and
must be kept with its checksum.
Exploring complementation experiments
Open Complementation Explorer from the Tools menu. It starts on yeast ACT1;
clear the search to see all 1,449 SGD records. One study, PMID:25999509, is a
bulk humanization screen and supplies 176 of them. Records per citing study
shows the ten studies with the most records and one bar for the other 703; the
bulk study is marked ◆ and drawn as an outlined bar. Untick Include
PMID:25999509 to hide its records: the status line drops to 1,273, and the
bars, graph and exports follow. The scope sentence above the bars states how
many records the study contributes under the current filters. Select a study
bar, or press Enter on it, to show only its records; the selected bar is marked
▶. Arrow keys move between bars. The toggle and the selected study are kept in
the address (#bulk=false, #study=PMID:…), so a link reopens the same view.
Download studies SVG saves the bars with their scope and snapshot metadata.
Selecting a connection shows a ChEMBL chip for each gene. Only a human HGNC partner is looked up; yeast genes read "Not evaluated", because activity is not transferred across orthology. A partner with collected records links to the Drug Evidence Explorer. The Drug evidence panel counts the human partners in the current set and names the ChEMBL release and the CC BY-SA 3.0 licence. If the drug index does not load, the panel says "Drug index unavailable." and the rest of the explorer keeps working.
Exploring OrthoDB groups
Open OrthoDB Group Explorer from the Tools menu. It starts on human TP53 at Hominidae. Choose a reference gene, then a taxonomic level; each level is a separate OrthoDB group returned for that gene, not a branch of a gene tree.
Members at each taxonomic level draws one bar per returned level, narrowest
first. TP53 has 7 levels, from 7 members at Hominidae to 796 at Vertebrata;
ACTB has 5, ending at 26,156 members in 5,797 assemblies at Eukaryota. Each
label gives the level, its taxon and the number of assemblies; the value is the
number of members. Bar lengths use a log scale. The search box narrows the bars
as it narrows the graph, and the scope sentence states the span, the filter and
how many levels have an assembly with more than one member. Select a bar, or
press Enter on it, to show that level's members in the graph; the shown level
is marked ▶ and arrow keys move between bars. Download levels SVG saves the
bars with their scope sentence and snapshot metadata, and the evidence JSON
carries the same counts under levels. OrthoDB groups carry no HGNC
identifier, so this explorer shows no drug data.
Drug Evidence Explorer: detailed workflow
The Drug Evidence Explorer is a bounded ChEMBL snapshot for eight selected human targets (listed on the page and in its manifest), their indexed orthologs and exact single-protein ChEMBL target matches. A target with no exact single-protein match, such as a tumour suppressor without a ChEMBL protein target, appears as a coverage row with no activity records. It helps inspect source records; it is not a target-ranking, drug-recommendation or clinical tool. Open the page's Detailed guide before starting if you need the definitions beside the interactive controls.
A reproducible review
- Choose a human target. Narrow to an exact assayed protein when needed; human and mouse proteins are separate ChEMBL targets.
- Apply the variant, assay type, measurement, assay-species, publication-year or compound/publication filters that fit the question. These controls filter activity rows. The mechanism-action graph filters the separate mechanism table and JSON export.
- Read the snapshot summary and target-coverage table. The snapshot keeps up to 100 IC50, Ki, Kd or EC50 activities with a pChEMBL value per mapped protein, highest pChEMBL first, plus the same measurements for that protein's mechanism compounds. A truncated window has additional source rows that are not displayed.
- Use the compound × protein matrix to see where sampled activity rows occur. Its compound count shows how many rows are in the current slice; scroll in either direction while the column headings and compound labels stay visible. Select a matrix cell or graph bar to filter; select a potency point to scroll to and highlight its matching activity row, then inspect the source context.
- Export JSON when coverage, identifiers and mechanism records are needed together. Export TSV for activity-row spreadsheet work; TSV does not contain the complete structured coverage and mechanism context.
Graph definitions and limits
- Measurement type by target counts sampled rows by exact target protein and source measurement label. IC50, EC50, Ki, Kd, inhibition and binding are separate categories and should not be pooled automatically.
- Assay species by target uses the reported organism of the experiment, not the target protein. Missing assay taxon is unknown, never inferred from target species.
- Variant annotation coverage distinguishes rows carrying a variant annotation from rows without one. Missing annotation does not mean the tested protein was wild type.
- Publication year counts rows by source document year in the bounded activity-ID sample. This cannot establish publication volume, historical trends or research attention.
- Mechanism action counts ChEMBL mechanism annotations by exact target and action. These are separate from measured activities and do not establish an indication or clinical effect.
- Comparable potency includes only exact, positive finite nM IC50, EC50, Ki or Kd records passing the source flag, confidence, relationship, single-protein, duplicate and validity checks. Points remain separated by target, assay species, variant, assay and endpoint. No average or combined score is calculated.
Every bar counts records returned in the current sample; it does not count independent experiments unless the source guarantees that distinction. “No exact target match” means the identifier lookup returned no exact single-protein target. A mapped target with zero sampled rows is a different state. Neither state means inactivity, absence of study or failed efficacy. Orthology is contextual information and never transfers a measurement between species. Target species, assay species, variant, relation and units must be read together.
The source snapshot, query limits, checksums, licenses and refresh workflow are
documented in docs/drug-evidence-pilot.md in the repository. ChEMBL data are
CC BY-SA 3.0; review attribution and reuse conditions before redistributing
derived datasets.
Study dossiers and molecular tools
Location: Study dossiers on BioWeaver. Open Study dossiers for the P2RX7, P2X4 and TRPA1 study selections. Each dossier keeps its own publication, licensing and assay context. These are selected study results, separate from the bounded ChEMBL explorer.
Inspect a measurement
- Choose a study and optionally filter by Compound or Assay readout.
- Select a plotted point or an Inspect button in the measurement table.
- Read the receptor species, assay host, protocol, replication and source in Measurement context. Assay host and receptor species are separate fields.
- Download measurements as JSON or TSV to retain the source values and context.
SD and SEM are different reported quantities. An absent uncertainty value produces no uncertainty whisker. The P2X4 BAY-1797 table/text discrepancy remains in the inspector and exports. These differences are not resolved by the plot.
Draw compounds and inspect a structure
- In Explore structures and chemistry, choose Draw compounds with RDKit. The drawing and descriptors are calculated from the dossier's supplied SMILES.
- Read the source and computed InChIKeys. Identifier mismatch means the supplied representation produces a different key; it is not an exact chemical identity confirmation. Both values remain available in the chemistry export.
- Choose a Deposited structure, then Load 3D structure. Mol* displays deposited model 1 from a checked coordinate snapshot.
- Use Focus … ligand to select the deposited ligand component. Drag to rotate the structure and zoom to inspect it. Reset structure view restores the view.
- Expand a ligand instance under Calculated protein–ligand contacts. Select a contact to focus its author chain and residue; the status reports the number of selected atoms. Download PLIP contact JSON for the full geometry and preparation record.
PLIP contacts are computed geometric classifications after coordinate preparation. They are not measured affinities. Author residue numbering in a structure can have a different meaning from a canonical UniProt sequence position. This viewer does not transfer activity to an ortholog, run docking, or establish selectivity.
P2X4 has no source SMILES or structure records in this selected study. Apo and other structures without a generated PLIP result say that analysis is unavailable; that is different from an analysed ligand with zero classified contacts. If a viewer fails, the assay table and its exports remain available.
Analysis importers and examples
After loading an example or file, Imported-data graphs shows three figures. OrthoFinder offers species membership totals, a group-size distribution and a group-by-species count matrix. InterProScan offers matches by analysis, matches per reported protein and covered sequence positions. Coverage merges overlapping and repeated intervals before dividing by the reported sequence length; it does not measure confidence or functional completeness.
The figures always describe the entire imported file, independently of table search. Use Next rows and, for the matrix, Next species to browse large inputs. Select a group or protein mark to open its records; arrow keys and Enter/Space also work. Download SVG exports the visible page; Download graph data (JSON) includes every summary row. Source checksums, version details and synthetic-example labels travel with both exports. The Software references panel links the tool papers and official documentation and offers a reference download. Cite the version actually used to produce your data.
Location: Analysis importers on BioWeaver. Open Import analysis results. The page reads local OrthoFinder membership tables and InterProScan match tables. It does not run either analysis program. Your uploaded file stays in the browser; reloading clears it.
Try the examples first
The Try an example cards load a small table without requiring a file upload. All three examples are synthetic demonstrations with invented identifiers and values. They are not published biological data or outputs of a software run.
| Example | What to inspect |
|---|---|
| OrthoFinder groups | Five groups across three illustrative species columns. Search for DEMO_OG0002, then select View members: one species has two members and another has no membership recorded. |
| OrthoFinder hierarchy | Two hierarchical groups. Open DEMO_N0.HOG0002 to see its parent orthogroup and gene-tree clade separately from species memberships. |
| InterProScan matches | Six match rows across three illustrative proteins. View matches for DEMO_PROTEIN_A shows overlapping location tracks. DEMO_PROTEIN_B has a repeated match, which is retained and flagged. |
Select Load … example to import the table immediately. Each card also has Download example TSV so you can examine the input format or import it through the regular form. The exact bundled bytes are recognized even if the downloaded file is renamed. The synthetic label appears in the result and stays in JSON and TSV exports. An edited file has a new checksum and is not automatically recognized as the bundled example; keep your own simulation or demonstration labels when editing it.
The examples leave software versions blank because no software run produced them.
DemoSignatures, DEMO_DOMAIN_*, DEMO_REGION_* and the example species labels
are invented. They do not identify real databases, signature entries or organisms.
Import your own result
- Choose Result format: OrthoFinder orthogroups TSV or InterProScan TSV.
- Choose the corresponding UTF-8 tab-separated Result file.
- Optionally enter Software version, Database or proteome release, and Run details. These fields are user-reported; the importer does not verify them.
- Select Import results. Read Import provenance and Interpretation and coverage before interpreting the table.
- Search an identifier and select View members or View matches. Use Previous and Next for tables longer than 100 rows.
- Download the result as JSON or TSV before reloading the page.
Changing the file, format or run metadata clears the previous imported result. Loading another example also clears it. Old asynchronous reads cannot overwrite a newer selection. No automatic identifier mapping to Loom genes is performed.
Supported inputs
| Format | Expected content | Common mistake |
|---|---|---|
| OrthoFinder membership | A header starting with Orthogroup, followed by species columns; comma-separated gene identifiers inside each cell. Empty cells stay empty. |
Supplying a gene-count or pairwise-ortholog table. Renaming a file does not convert its contents. |
| OrthoFinder hierarchy | HOG, OG, Gene Tree Parent Clade, then species membership columns. |
Treating the three metadata columns as species or treating group membership as one-to-one orthology. |
| InterProScan | Headerless TSV with 11, 13, 14 or 15 columns; match locations use 1-based inclusive protein coordinates. | Adding a spreadsheet header, converting tabs to commas, or supplying a match end beyond the reported protein length. |
The input limit is 5 MiB, 25,000 data rows, and 100,000 OrthoFinder memberships. Large native analysis outputs must be deliberately subsetted before importing; record that selection in Run details. The importer rejects malformed files as a whole instead of presenting a partial result as successful.
Exports and interpretation
JSON retains original row fields, source line numbers, source SHA-256, missing values, run metadata and the synthetic-example label when recognized. TSV repeats source provenance on each data row and protects cells that resemble spreadsheet formulas. Use JSON when you need exact literal identifiers or provenance for an empty imported result.
Search changes the displayed table, not the download: both export buttons include the entire imported result. The processed TSV export has additional provenance columns and is not an OrthoFinder or InterProScan input file. Use Download example TSV when you want a raw example that can be imported again.
An empty membership cell does not prove gene absence. InterProScan match tables do not enumerate all searched proteins or establish that every analysis succeeded. Scores remain method-specific, repeated rows remain visible, and missing scores are not zero. Group membership and signature similarity do not transfer drug activity.
When something goes wrong
| Message or symptom | Next step |
|---|---|
| Example checksum mismatch or download failed | Reload and retry. The example is not displayed if its bytes differ from the bundled receipt. |
| Wrong number of TSV columns | Use the original tab-separated output. Remove spreadsheet-added headers only if the source format is headerless. |
| Duplicate orthogroup membership | Check whether files from different runs or hierarchy levels were combined. Import them separately. |
| Conflicting sequence length or MD5 | Separate records that use the same protein identifier for different sequences. |
| Coordinates outside the sequence | Check the original InterProScan output and sequence length; do not clip positions to make the file pass. |
| File exceeds a limit | Export a documented subset from the native analysis, retaining enough source information to describe the selection. |
| Results disappear after editing metadata | This is expected. Import the selected file again so the displayed provenance matches the current fields. |