Discovery Program
Arclight is an AI-native commercial development discovery platform. Upload a patient cohort and clinical question — autonomous agents surface non-obvious, evidence-ranked programs from anchor populations through IND roadmap.
Built for teams who need discovery beyond what expert-gated review surfaces.
Portfolio queue
Every program, one view
Sort by confidence, recency, or phase. Act Now · Too early · Crowded.
Live evidence
Arclight does not rely on cached demos or frozen snapshots. Each agent step pulls fresh evidence from authoritative biomedical sources.
PubMed
Literature & preprints
ClinicalTrials.gov
Trial landscape
Patents
IP & prior art
The challenge
Built for: BD & licensing · Discovery leads · CSO office · Innovation teams
The gap
Teams evaluate a handful of opportunities over months. Assessments are static, expensive, and anchored on familiar targets. The patient subgroup inside a heterogeneous indication — and the mechanism linking two unrelated conditions — stay invisible.
Arclight starts from your cohort and clinical question, not your existing search taxonomy.
The approach
Autonomous agents query public biomedical APIs, rank evidence, and narrow hypotheses — revealing opportunities that committee-driven review structurally misses. Every step remains visible with sources and reasoning.
Launch a full discovery session and review the audit trail yourself.
The outcome
Commercial outcomes. Market-sized anchors, Act Now actionability zones, and IND-ready regulatory roadmaps support portfolio ROI decisions.
Discovery efficiency. Autonomous pipeline compresses identification time; undruggable registry eliminates repeated dead-end spend.
From cohort upload to ranked program — in one continuous pipeline.
How a discovery program flows
Cohort
Upload patients + clinical question
Anchors
Define biology & resistance populations
Expert domains
Mine cohort + literature adjacencies
Funnel
50 → 20 → 3 evidence-ranked hypotheses
Portfolio
Ranked programs with confidence tiers
Anchor methodology
After cohort upload, Arclight defines biology and resistance anchor populations — falsifiable subject–relationship–outcome claims tied to your query-defined patients. Market sizing runs on each anchor before hypotheses multiply.
Finds the patient subgroup hidden inside a heterogeneous indication — not the indication label alone.
Population-first, not mechanism-first.
Biology anchor
Anti-PD1 non-responders with elevated TGF-β drive resistance via immunosuppressive stroma.
Resistance anchor
TGF-β signaling in the tumor microenvironment limits checkpoint response.
Expert domain discovery
CD1 mines co-occurring scientific domains from your patient cohort outside the parent therapeutic area. CD2 scans PubMed and target databases for disease, organ, and molecular associations. The merger produces expert domains that feed cross-context hypothesis seeds.
Surfaces mechanisms connecting conditions your taxonomy treats as unrelated.
Surfaces adjacencies your team wouldn't search.
Domain discovery pipeline
CD1 · Cohort patterns
Recurrence from patient data
CD2 · Literature
PubMed & target associations
Hypothesis funnel
Phase 1 runs a progressive selectivity filter: 50 falsifiable association hypotheses, 20 promoted with mechanistic chains, exactly 3 ranked by anchor fidelity, evidence strength, and selective intervention feasibility.
Evaluate breadth in one session — not a handful of opportunities over months.
Three survivors, ranked by anchor fidelity and selectivity — not taxonomy convenience.
Progressive selectivity filter
Association filter
Broad exploration, anchor-linked
Causation filter
Mechanistic promotion
Selectivity rank
Intervention-ready survivors
Phase 2
After funnel convergence, Phase 2 screens druggability across small molecule, biologic, and ADC, then assembles drug path, IP/FTO, TPP, risk scoring, and an IND regulatory roadmap — decision-grade output, not a slide.
Compresses months of cross-functional assessment into one autonomous Phase 2 run.
Phase 2 development path
Step 1
Druggability screen
Small molecule, biologic, and ADC gate
Step 2
Drug & IP
Modality path, freedom to operate
Step 3
TPP blueprint
Falsifiable product claim
Step 4
IND package
Preclinical, CMC, tox, and regulatory roadmap
Undruggable registry
When three-modality druggability screening fails, Arclight records the target and blocks it in future programs — globally or session-scoped — with reasoning and alternate intervention paths preserved.
Cuts repeated identification cost on targets already proven undruggable.
12
Registry entries
8
Unique targets
5
Global blocks
3
Rescan eligible
Registry at a glance
Targets that fail three-modality screening are blocked in future programs. Global and session-scoped entries with alternate intervention paths.
Portfolio
Every discovery program lands in a sortable portfolio queue with actionability zones — Act now, Too early, Crowded — and discovery confidence scoring for commercial prioritization.
Portfolio ROI decisions grounded in confidence tiers, not static point-in-time memos.
Trust
Every agent step, evidence card, and source citation is visible. Humans review reasoning — agents do the searching.
Agents run the pipeline; humans review evidence, sources, and reasoning.
Live agent activity stream
Watch agents query PubMed, trials, and patents in real time.
Evidence cards with citations
Every claim links to its public source.
Program trust scoring
Funnel coverage and evidence depth in one confidence tier.
Regulatory provenance
Act Now programs get FDA/EMA-ready assembly packages.
