Corroborate a GO/NO-GO with peer-reviewed literature, surface the pathway entities behind it, and replay every result as a Proprietary Truth.
Support Rationale: from a BVCT result to a cited graph
A BVCT result tells you the GO or NO-GO verdict, the odds, and the commercial prediction. Support Rationale is the evidence layer that sits underneath that verdict: it searches the published literature for papers that corroborate the call and, where it can, lifts the genes and proteins those papers name back onto your knowledge graph so the rationale becomes something you can read, click, and trace.
The feature lives at the seam between the two modules. You produce a result in the Validator, press Visualize Rationale on the result summary, and the platform hands the result over to the Generator, which renders the drug → indication relationship as a knowledge-graph edge and then enriches it with citations. Nothing here is hand-authored copy — every paper and every pathway entity is retrieved at the time you ask for it.
Two modules, one flow. Support Rationale is triggered from the Validator (Visualize Rationale) and rendered in the Generator. You do not need to manually rebuild anything — the Generator picks up the handed-over result, builds the graph, runs the publication search, and adds pathway nodes automatically.
What gets carried across
When you click Visualize Rationale, the Generator receives the result's key fields and uses them both to draw the graph and to seed the literature search:
Field
Role in the rationale
Trial Drug
One endpoint of the edge; the primary search term.
Trial Indication
The other endpoint; matched to a disease node in the graph.
Commercial Prediction
Context for relevance scoring of papers.
Key Clinical Finding
Sharpens the search toward the mechanism that drove the verdict.
Key Success Factor
Further focuses literature on the benefit rationale.
GO / NO-GO, odds, category
Stored on the resulting edge so the verdict travels with the graph.
1
Produce a result, then Visualize Rationale
In the Validator, open a completed BVCT and press Visualize Rationale on the result summary header. The platform navigates to the Generator carrying the drug, indication, verdict, and odds.
Tip: No matching drug node yet? The Generator will create a new hypothesis space named after the BVCT ID and place a purple hypothetical drug node automatically — you do not have to pre-build the graph.
2
Watch the graph assemble
A building badge appears while the Generator locates (or creates) the drug and indication nodes, draws the connecting edge, and kicks off the literature search. GO edges render in purple/green; NO-GO edges render in red with an X marker instead of an arrowhead.
3
Read the cited rationale
Click the edge to open its annotation. The top corroborating publications are listed there, each with title, authors, journal, year, a relevance score, and a one-line note on why it supports the result. Pathway-entity nodes (if any were found) branch off the indication, each tagged with its source paper.
Publication search: finding corroborating papers
The literature search is the heart of Support Rationale. It runs an AI literature-research agent (Gemini with live web search) against the result's drug + indication and returns the top 10 most relevant peer-reviewed publications that corroborate the call, sorted by relevance score (highest first).
What the agent looks for
Publications that directly discuss the trial drug in the context of the trial indication.
Clinical-trial results, meta-analyses, and systematic reviews are prioritised.
Papers that support the drug's mechanism of action for that drug–indication pair.
Papers that discuss the clinical-benefit rationale behind the prediction.
Recent work (last ~10 years preferred) plus seminal older papers.
What each result carries
Field
Description
Title
Full publication title.
Authors
First three authors et al.
Journal & year
Source venue and publication year.
URL
A link to the paper (PubMed, journal site, or DOI).
Relevance score
1–10, how strongly the paper corroborates the BVCT result.
Relevance summary
A 1–2 sentence note on why the paper is relevant.
The list is formatted into the edge annotation as a ranked block — Top N Corroborating Publications — so the strongest evidence sits at the top. Results are filtered to keep only entries that have a real title and URL, then capped at ten.
Resilient by design. If the primary model is overloaded the service retries with exponential backoff and automatically falls back through a priority list of models before giving up. A transient outage produces a single placeholder row ("Publication search temporarily unavailable") rather than a hard error — just press Visualize Rationale again later.
When the indication matches more than one disease
Indication matching is keyword-based and apostrophe-insensitive, so a result for "Early Alzheimer's Disease" will find any disease node whose name contains "alzheimer". If several disease nodes match and no drug node exists yet, the Generator pauses and shows an indication selection dialog. Pick the disease you mean, confirm, and the search resumes against that node.
1
Pick the right indication node
Select the disease node the result refers to and confirm. The Generator then creates the drug edge to that exact node and runs the publication search.
Tip: If none of the offered nodes is correct, cancel — the rationale is discarded cleanly and nothing is added to the graph.
Evidence, not proof. Corroborating papers are retrieved to support a model-based verdict; they do not certify it. Always read the relevance summary and open the source before treating any citation as authoritative. BVCT outputs are decision-support analyses, not investment advice.
The Truths tab: replaying validated results
Every BVCT you validate is captured in the Generator's left sidebar under the Truths tab, in a section labelled PROPRIETARY TRUTHS. This is your running ledger of GO/NO-GO calls — each one a path you can re-open in the 3D graph whenever you need to revisit the rationale behind it.
Open the Generator, then select the Truths tab in the left sidebar:
Each validated BVCT shows as a row: a GO or NO-GO pill, the drug → disease pair, the node count, and the odds percentage. Until you have run a validation, the section reads "No BVCT results yet. Validate hypotheses to see GO/NO-GO truths."
Tip: The GO/NO-GO pill carries a tooltip: "Computed from the source of ex-ante truth: data.bioinvestgpt.com." That prospective track record is published at data.bioinvestgpt.com.
2
View a truth in the graph
Use the View in Graph control on a row to load that truth into the 3D view. The selected path is highlighted — green glow for GO, red for NO-GO — and its corroborating paper nodes and pathway entities are reconstructed around the disease.
Deep-linking a single truth
A specific truth can be opened directly by BVCT ID. The deep link routes straight to the Truths tab and auto-loads that path into the graph:
When a truth is loaded, the BVCT verdict travels with it. Click the drug → indication edge and expand its annotation: the BVCT Validation block is shown first and prominently, with the GO/NO-GO result, the badge category, and the odds rendered as colour-coded pills (green for GO, red for NO-GO) above the corroborating-publication list.
Truths persist. Validated results are retained across sessions and re-synced from the backend, so your Proprietary Truths list reflects every BVCT you have run, not just this session's. Newly validated hypotheses trigger a recompute so the list stays current.
Pathway entities & paper nodes on the graph
Beyond a flat citation list, Support Rationale can lift the biological actors named in the corroborating literature back onto the knowledge graph. These pathway entities — genes, proteins, receptors, and pathways — are extracted from the papers and rendered as nodes connected to the indication and to the papers that mention them, turning the rationale into a small mechanistic sub-graph.
What a pathway entity is
Property
Meaning
Name
The gene/protein symbol (e.g. PDCD1, CTLA4, PD-L1).
Type
One of: gene, protein, pathway, or receptor.
Role
A one-line description of the entity's role in the drug–disease pathway, drawn from the paper context.
Source paper
A link back to the specific publication the entity was extracted from.
How they land on the graph
Enrichment is additive and best-effort — it only adds, never removes, and silently skips anything it cannot resolve:
Each entity is matched by name against existing knowledge-graph nodes. If a match is found, the node is made visible (and enlarged as a path endpoint) and an edge is drawn from the indication to it.
The edge is annotated with the entity's role and a reference to its source paper — title, journal, year, and URL.
Pathway edges inherit the verdict colour: green for GO, red for NO-GO.
Entities whose symbol is not in the graph are skipped without error — an absent gene never breaks the build.
Paper nodes and how they connect
When you re-open a truth via View in Graph, the corroborating papers themselves are added as purple publication nodes arranged in a ring around the disease. Each paper node carries its title, journal source, authors/year/impact-factor summary, and URL. Papers connect two ways:
To the pathway entities they mention, via "corroborates" edges — alias-aware, so PD-1 resolves to PDCD1 and HER2 to ERBB2.
To the disease node itself, via "evidence for" edges.
Read the graph like a story. Drug → indication is the claim; indication → pathway entity is the mechanism; paper → entity ("corroborates") and paper → disease ("evidence for") are the citations. Click any edge to read the underlying annotation and follow the source link.
Why do some results show no pathway entities?
Pathway enrichment only runs when high-impact biological pathway papers were found and entities could be extracted from them. If the literature for a drug–indication pair is thin, or the named genes are not present in the knowledge graph, you will still get the corroborating-publication list on the edge — just without the extra entity nodes.
Are paper nodes saved or re-fetched each time?
When you View in Graph, the platform first loads the saved rationale for that BVCT (its stored pathway papers and entities); if none is on file it falls back to any pathway data carried with the result. Previous publication nodes are hidden before new ones are drawn, so re-opening a truth does not pile up duplicates.
Symbol matching is exact-by-name. Pathway entities only attach to graph nodes whose names match the extracted symbol (alias resolution applies for paper-to-entity links). A correctly cited gene that simply is not in your current graph will appear in the paper annotation but not as its own node — that is expected, not a bug.
BioinvestGPT BVCT Platform User Guide — Publication Search & Rationale. BVCT outputs are model-based decision-support analyses, not investment advice. Prospective track record: data.bioinvestgpt.com.
