Data Sources & Methodology

Source: We parse the FDA Drugs@FDA bulk data files which contain all FDA-approved drug applications (NDAs, ANDAs, BLAs) since 1939.

Enrichment: Drug labels from DailyMed are parsed to extract indications, mechanism of action, targets, warnings, and adverse reactions.

Normalization: Brand names and generic names are normalized. Company names are standardized across acquisitions and name changes.

Source: We maintain a local index of ~147,000 clinical trials from ClinicalTrials.gov for dashboard analytics, covering interventional Phase 1-4 trials started from 2008 onwards.

Live queries: The Trials Explorer queries ClinicalTrials.gov in real-time, searching all 440,000+ trials in the full CT.gov database - not limited to our indexed subset.

Filtering: Our index focuses on interventional drug trials in Phase 1-4 (excluding device, behavioral, observational, and non-phased studies). The Trials Explorer live search has no such filter and returns all matching studies.

Manual curation: We maintain a hand-curated taxonomy that maps FDA indication text and clinical trial conditions to standardized therapeutic areas.

Coverage: See our taxonomy dashboard for current mapping coverage and completeness metrics.

Therapeutic areas: Oncology, CNS, Cardiovascular, Metabolic, Infectious Disease, Immunology, Respiratory, Rare Disease, and more.

Method: We search SEC EDGAR full-text for PDUFA-related keywords in 8-K, 10-K, and 10-Q filings.

Keywords: "PDUFA", "Prescription Drug User Fee Act", "FDA target action date", "FDA goal date"

Limitations: Not all companies disclose PDUFA dates. We supplement with manual entries for known upcoming decisions.

Orange Book: Contains patents and exclusivity for small molecule drugs (NDAs). Updated monthly by FDA.

Purple Book: Contains biosimilar interchangeability and reference product designations for biologics (BLAs).

Limitation: Many biologics are not listed in the Orange Book. Patent cliff analysis focuses on drugs with Medicare Part D spending data.

Approved drugs: We read targets from the FDA label on DailyMed, in the "Mechanism of Action" and "Clinical Pharmacology" sections, using AI-assisted extraction. We normalize them to standard HGNC gene symbols (like GLP1R, EGFR, TNF) where possible. Cell and gene therapies take their target from our curated cell and gene therapy list.

Coverage: 723 targets across 1,721 approved drugs. These make up the target pages. Not every label states a target, so older drugs and some generics have none.

Drugs still in trials: The target pages list approved drugs only. Investigational drugs get a target or mechanism on other pages, each with its own method:

• Disease landscapes and company portfolios (heatmaps). We match each trial program to a mechanism in layers: our curated rulebook first, then public molecular-target data. Only when neither resolves a sponsor's code name do we ask an AI model, and its answer is checked against the sponsor's pipeline page; an answer the sponsor contradicts is dropped. The page marks every AI-assigned mechanism (✅ matches the sponsor's page, ⚙️ not yet checked) and explains the method in full.
• Mechanism pages start from a curated roster for each mechanism and add trial programs whose target matches. Some of those targets were assigned by an AI model, and these pages do not yet mark which.
• Emerging drugs: 18 investigational drugs whose targets we checked by hand.

Limitation: Investigational drugs are covered only for the diseases and companies we have built heatmaps for, so these pages are not a complete list of every program against a target. Where a sponsor has not disclosed a mechanism, the heatmaps show "not yet disclosed" rather than a guess.

Source: Open Targets Platform integrates genetic associations from GWAS studies, rare disease genetics, and functional genomics.

What we show: For each drug target, we display diseases with genetic evidence linking mutations in that gene to disease risk (e.g., GLP1R variants associated with Type 2 diabetes).

Why it matters: Drugs targeting genes with strong genetic evidence have historically shown higher clinical trial success rates. Genetic validation de-risks target selection.

Scoring: Open Targets provides association scores (0-1) based on evidence strength, study size, and statistical significance.

Where AI helps: Several data pipelines use AI-assisted extraction. Drug targets are extracted from DailyMed labels using LLM analysis. On the heatmaps, an AI model proposes a mechanism only for trial code names that our rules and public target data cannot resolve, and the page marks it. PDUFA dates are extracted from SEC EDGAR full-text using LLM-aided keyword analysis. Brief research involves AI-assisted aggregation of primary sources — FDA approval letters, SEC filings, peer-reviewed publications, press releases — and AI-assisted draft generation.

Where it doesn't go: Editorial judgment, source verification, and the final voice on every published Brief are mine. Every specific number, date, trial result, or regulatory decision cited in a Brief is verified against the primary source before publishing. Cross-trial comparisons, head-to-head positioning, and competitive claims are written and stress-tested by hand.

Why this is transparent: AI accelerates the work — research aggregation, data pipeline maintenance, draft scaffolding — but the editorial product is mine. If you find an error, it's mine to fix. Let me know.