Biological Evidence Analysis
A mechanism class with one Phase 2b win, two recent Phase 2 failures, and zero peer-reviewed publications on the lead program
An investor asked whether a private-stage PD-1 agonist company was differentiated from its failed predecessors. The pitch deck said yes. The evidence said something more complicated.
The Investment Question
PD-1 checkpoint inhibitors — Keytruda, Opdivo — generated $60B in 2025 oncology revenue by blocking the PD-1 brake on immune cells. The side effects tell the other story: thousands of cancer patients developed autoimmune conditions. Colitis. Thyroiditis. Type 1 diabetes.
The thesis is elegant: if blocking PD-1 breaks immune tolerance, activating PD-1 should restore it. The $168B autoimmune market is the prize. But elegant biology has a body count in this mechanism class. Eli Lilly terminated peresolimab Phase 2b in rheumatoid arthritis in October 2024 on overall benefit/risk grounds. AnaptysBio's rosnilimab failed Phase 2 in ulcerative colitis in November 2025 — 7% clinical remission, no better than placebo at Week 12, trial discontinued.
The company claims their bifunctional design solves the problems that killed the first-generation programs. That's the claim every second-generation company makes. We were asked to evaluate whether the public evidence supports it.
What the Evidence Shows
The target is validated. The approach is not.
We analyzed all 92,721 nivolumab adverse event reports in the FDA FAERS database and computed proportional reporting ratios for 12 autoimmune-relevant events against the full FAERS background (~20M reports). Eleven of twelve cleared the PRR > 2.0 threshold with the lower 95% confidence bound above 1.0 — the pharmacovigilance signal is mechanistic, not noise. GTEx tissue expression data independently predicts the adverse event pattern. GWAS hits at the PDCD1 locus for hypothyroidism and atopic dermatitis provide genetic convergence.
Three independent evidence layers — pharmacovigilance, expression, genetics — all point to PD-1 as a legitimate autoimmune target. This is not a theoretical mechanism.
The class track record is worse than the pitch decks suggest.
We mapped global PD-1 agonist clinical programs across ClinicalTrials.gov, SEC filings, and company disclosures. The scorecard at the time of writing: one Phase 2b win (rosnilimab in RA, AnaptysBio), two recent Phase 2 failures (rosnilimab in UC, peresolimab in RA), and an early Celgene/BMS program (CC-90006) discontinued in Phase 1. Translation of PD-1 agonism into clinical benefit has a thin track record.
Peresolimab's Phase 2b termination is the diligence signal that matters. Lilly cited overall benefit/risk — not a specific safety event — but discontinuing a Phase 2b trial in this disease area is the kind of decision a sponsor only makes when the numbers don't cohere. Any new entrant must demonstrate a differentiated efficacy and safety profile. Phase 1 PD biomarker data is the minimum threshold for further engagement.
The lead program has no peer-reviewed evidence.
We verified 23 claims from the company's public materials against independent databases. The lead program's entire evidence base is conference posters — ACR 2024 and 2025. No journal publications. No independent replication. No published binding kinetics.
The company's earlier program (not the lead) does have two PNAS papers with credible data. That program validates the team's scientific capability. It does not validate the lead molecule.
The Verdict
Monitor, do not commit.
The biological evidence supports PD-1 agonism as a mechanism. It does not support this company at this stage. The gap between a validated target and a validated molecule is where most biotech capital goes to die. Phase 1 pharmacodynamic biomarker data — expected late 2026 — is the de-risking catalyst. Selective T-cell suppression with preserved Tregs and no malignancy signal would differentiate this program from every failed predecessor. Anything less, and the 25% class success rate is the base case.
Why This Exists
You can ask a general-purpose AI about PD-1 agonism. It will give you a plausible summary. It will not pull live FAERS counts for Type 1 diabetes under nivolumab, then compute disproportionality against the FAERS background, because it never queried the FDA FAERS database. It will not retrieve the actual NCT05516758 trial record for peresolimab Phase 2b — enrollment, status, termination date, sponsor statement — and quote it back to you, because it cannot read live regulatory filings. It will not cross-reference GTEx tissue expression at the PDCD1 locus against the pharmacovigilance signal pattern to confirm mechanistic convergence, because it has no access to those databases.
Our analytical stack integrates 2,300+ biomedical research tools across 500+ data sources — FDA pharmacovigilance, ClinicalTrials.gov, PubMed, OpenTargets, GTEx, GWAS Catalog, ClinVar, FAERS, UniProt, STRING, AlphaFold, Reactome, KEGG, ChEMBL, DrugBank, GnomAD, ENCODE, and hundreds more. This report used 40 of them and made 597 API calls. A different target might require a different 40. The point is not the count — it is that every finding is queried live and traceable to a specific database, call, and timestamp. Nothing is synthesized from training data. Everything is verified from source.
The other gap is judgment. AI will hedge. This report says “monitor, do not commit” and names the specific Phase 1 biomarker readout that would change the recommendation. That opinion comes from 25 years of evaluating drug programs — at Novartis NIBR, during Shire's $32B acquisition diligence, and across 500+ compounds evaluated in search and evaluation roles. The system gathers the evidence. The judgment is mine.
Evaluating a target or mechanism?
I produce independent biological evidence analysis for biotech investors — pharmacovigilance, competitive landscape, claims verification, and adversarial review. The deliverable is a signal report with full data provenance. No black boxes.
Background: 25 years in biotech — drug discovery at Novartis, $32B M&A diligence at Shire, AI platform development as a founder. I read the biology, not just the pitch deck.
chris@bigbio.ai