Ask a research question.
Get an evidence-backed answer.
Scout orchestrates hundreds of specialized AI agents to run multi-step scientific investigations autonomously. One prompt. Full provenance.
One prompt. A complete scientific investigation.
FOLR1-Targeted CAR-T: A Tumor-Selective Target with a Defined Safety Roadmap
Ensemble synthesis of expression, clinical precedent, and toxicity evidence — ovarian-cancer program
1Original Research Question
Evaluate the on-target/off-tumor safety of a FOLR1-directed CAR-T for ovarian cancer by integrating GTEx bulk RNA-seq, CELLxGENE single-cell expression, clinical precedent, and structured toxicity evidence — then produce a tissue-level safety roadmap.
2Key Finding
In a single prompt, Scout integrated four independent evidence streams into a decision-ready development package with explicit confidence and next steps.
Evidence sources: GTEx bulk RNA-seq (54 tissues); CELLxGENE single-cell atlas; UniProt P15328; DepMap; ClinicalTrials.gov; 6 PubMed studies.
Tool chain: Expression retrieval → single-cell profiling → composite risk scoring → ensemble synthesis (2 runs).
Confidence: High for tumor-selectivity and the hematopoietic window; Moderate for the complete organ profile (7 tissues remain coverage gaps; no FOLR1 CAR-T clinical safety data yet).
Still to confirm: Direct tumor-vs-normal protein-density (IHC) for lung and renal tubule; primary-cell functional CAR-T testing.
3Normal-Tissue Expression (GTEx, 54 Tissues)
| Tissue | FOLR1 (median TPM) | Role in the roadmap |
|---|---|---|
| Normal ovary | 0.140 | FAVORABLE defines the window |
| Lung | 109.863 | Priority diligence — direct tissue-density confirmation |
| Minor salivary gland | 79.135 | Confirm with IHC |
| Kidney cortex / medulla | 50.5 / 52.8 | Priority diligence — proximal-tubule assessment |
| Thyroid | 33.956 | Routine monitoring |
| Fallopian tube | 5.845 | Relevant to HGSOC biology |
4Single-Cell Resolution
Single-cell profiling confirms FOLR1 is restricted to specific epithelial/secretory populations, with hematopoietic/immune compartments effectively negative.
| Cell type | Expression mean | Cells expressing |
|---|---|---|
| Pancreatic ductal cells | 17.622 | 69/389 (17.7%) |
| Respiratory club cells | 0.593 | 63/209 (30.1%) |
| Liver cholangiocytes | 0.213 | 49/474 (10.3%) |
| Class-switched memory B cells | 0.075 | 4/638 (0.6%) |
UniProt P15328 (GPI-anchored receptor) and DepMap non-essential classification (mean dependency −0.0372) are consistent with tolerance of modulation.
5Clinical Precedent
The FOLR1-directed ADC mirvetuximab soravtansine provides a mature playbook: its most common events are ocular, largely low-grade and mostly reversible, and the evidence suggests they are substantially payload-mediated — so they may not carry over to CAR-T.
| Event | Rate | Detail |
|---|---|---|
| Blurred vision | 43% (integrated) | 71% resolved; predictable onset (~6 wks) |
| Keratopathy | 29–48% | 66% resolved; grade ≥3 only 5% |
6Tissue-Level Safety Roadmap
| Tissue | Signal | Diligence action |
|---|---|---|
| Normal ovary | 0.140 TPM | FAVORABLE window confirmed |
| Lung / kidney | High bulk expression | FOCUS tissue-density + functional testing |
| Thyroid / salivary gland | Moderate–high bulk | CONFIRM targeted IHC |
| Ocular surface | ADC precedent (payload-linked) | MONITOR established surveillance |
| Bone marrow / blood | Effectively zero | CLEAR validated window |
7Path Forward
- Confirm the window: comparative tumor-vs-normal FOLR1 protein density (lung, renal tubule).
- Engineer selectivity: affinity/avidity tuning and activation-threshold control.
- Differentiate modality: compare the same binder in ADC vs CAR-T formats.
Go/no-go: tumor-selective activation with minimal activity against primary normal cells — a defined, achievable bar this evidence base supports pursuing.
Osimertinib versus Lazertinib: Two Leading First-Line EGFR-TKIs
Cross-trial synthesis of pivotal phase 3 efficacy and safety in EGFR-mutant NSCLC
1Original Research Question
Compare osimertinib versus lazertinib as first-line therapy in EGFR-mutant (exon 19 deletion / L858R) advanced NSCLC, synthesizing pivotal phase 3 efficacy, safety, and overall-survival evidence.
2Key Finding
Scout extracted and aligned the pivotal phase 3 evidence for both agents across two independent runs.
Evidence sources: PubMed (FLAURA, LASER301); ClinicalTrials.gov (250 vs 34 trials); ChEMBL; openFDA labels.
Tool chain: Literature retrieval → structured cross-trial extraction → indirect synthesis (2 runs).
Confidence: High for individual trial values and PFS-HR similarity; Moderate for cross-trial safety (differing populations/grading); Low for any claim of true superiority.
Still to confirm: Mature LASER301 OS data; a formal IPD network meta-analysis (shared gefitinib comparator).
3Head-to-Head Efficacy
| Endpoint | Osimertinib · FLAURA | Lazertinib · LASER301 |
|---|---|---|
| Patients | 556 | 393 |
| Median PFS | 18.9 vs 10.2 mo | 20.6 vs 9.7 mo |
| PFS HR | 0.46 (0.37–0.57) | 0.45 (0.34–0.58) |
| ORR | 80% | 76% |
| Median DoR | 17.2 mo | 19.4 mo |
| Grade ≥3 AE | 34% | 40% |
| AE discontinuation | 15% | 15% |
4Overall Survival
| Agent | OS result | Status |
|---|---|---|
| Osimertinib | 38.6 vs 31.8 mo; HR 0.80 (0.64–1.00); P=0.046 | Mature, significant |
| Lazertinib | 18-mo OS 80% vs 72%; HR 0.74 (0.51–1.08) | Immature at interim |
5Decision Guidance
- Today: osimertinib is the evidence-backed first-line choice (mature OS + 6+ years real-world experience).
- Watch: lazertinib PFS is fully competitive; maturing LASER301 OS will complete the picture.
- Next step: formal IPD network meta-analysis via the shared gefitinib comparator.
Antibody Developability & Immunogenicity: Favorable Profile, One Verification Target
Germline confirmation, developability screen, and HLA class II epitope triage for a human IgG1 candidate
1Original Research Question
Assess the anti-drug-antibody (ADA) risk and developability of a human IgG1 candidate via germline analysis, HLA class II epitope prediction, and a developability screen — and identify precise, testable follow-ups.
2Key Finding
Scout ran a full developability and immunogenicity triage and returned a candidate that is strong on fundamentals, with one precisely-located item to confirm.
Evidence sources: VH/VL sequence; germline databases; AFND HLA population frequencies; developability panel.
Tool chain: Germline assignment → MHC-II epitope prediction (internal custom model) → population coverage → developability + spatial aggregation propensity → deterministic reconciliation.
Confidence: High for germline & developability (aggregation/stability); Indeterminate for population-level ADA (allele coverage recovered 0/26 — flagged, not hidden).
Still to confirm: Experimental HLA-DRB1*01:01 binding + CD4⁺ T-cell activation for the KLLIYSASFLYSGVP region; rerun MHC-II with full allele coverage.
3Developability — Strong Fundamentals
| Property | Result | Read |
|---|---|---|
| Germline (heavy / light) | IGHV3-66*01 / IGKV1-39*01 | Fully human |
| Spatial aggregation propensity | −80.0 (threshold 10) | Favorable |
| Aggregation hotspots (≥1.0) | 0 | None |
| Destabilizing residues | 0 | None |
| Chemical-motif liabilities | 12 (2 high, 2 mod, 8 low) | Routine forced-degradation checklist |
4Immunogenicity — A Precise, Testable Next Step
The MHC-II triage flagged incomplete allele coverage rather than over-claiming, and surfaced a single CDR2-adjacent light-chain peptide worth confirming. For HLA-DRB1*01:01 the peptide KLLIYSASFLYSGVP (light-chain positions 45–59) was a dual-method call at 49 nM.
5Recommended Path
- Confirm: experimental DRB1*01:01 binding + CD4⁺ T-cell activation for the flagged region.
- Standard QC: forced-degradation on the two high-severity motifs.
- If confirmed benign, the fully-human, low-aggregation profile supports advancement.
Structure-Guided Factor Xa Inhibitor Design: A Prioritized, Drug-Like Candidate Panel
De novo generation on the 2W26 pocket with docking, drug-likeness, and synthetic-accessibility triage
1Original Research Question
Generate developable small-molecule Factor Xa inhibitors from a 2W26 structure-guided workflow, and evaluate candidate quality with docking, synthetic-accessibility, and drug-likeness filters.
2Key Finding
From a single prompt, Scout ran a full structure-based generation and triage pipeline against the Factor Xa pocket and ranked a shortlist for assay.
Evidence sources: Factor Xa structures 2W26 & 2P16; PubChem; BindingDB Factor Xa reference set.
Tool chain: Pocket2Mol generation → AutoDock Vina docking (2 receptors) → RDKit SA + descriptors → PubChem identity/similarity → triage.
Confidence: Computational only. Docking scores and QED/SA are in-silico proxies, not measured affinity or synthesizability. 32/55 were exact public matches; lig_017 was a similar-only match (potentially differentiated).
Still to confirm: Purified Factor Xa concentration-response assays; identity/pose confirmation; novelty & selectivity vs related serine proteases.
3Docking Performance (AutoDock Vina)
| Metric | 2W26 | 2P16 (cross-receptor) |
|---|---|---|
| Ligands docked | 55 | 54 (paired) |
| Best predicted affinity | −8.338 kcal/mol | −8.313 kcal/mol |
| Median predicted affinity | −5.887 kcal/mol | — |
| Poses / exhaustiveness / failed | 5 / 32 / 0 | — |
4Prioritized Candidate Panel
| Candidate | 2W26 dock | QED | SA | MW (Da) | Note |
|---|---|---|---|---|---|
| lig_017 | −8.338 | 0.895 | 1.78 | 255.3 | LEAD best dock + drug-likeness; similar-only match |
| lig_041 | −7.234 | 0.738 | 1.40 | — | High QED, very synthesizable |
| lig_016 | −7.678 | 0.618 | 2.95 | 172.2 | Solid assay candidate (exact match) |
| lig_030 | −7.717 | 0.561 | 3.14 | 160.2 | Solid assay candidate (exact match) |
5Recommended Path
- Advance the panel to purified Factor Xa concentration-response assays.
- Confirm identity/pose for lig_017; assess novelty vs the closest public match.
- Follow through: retrosynthesis, solubility/stability, selectivity vs related proteases.
Direct Factor Xa Inhibitors as Alternatives to Rivaroxaban — Apixaban Leads
Evidence-ranked comparison across VTE prophylaxis, AF stroke prevention, and DVT/PE treatment
1Original Research Question
Systematically examine approved direct Factor Xa inhibitors as alternatives to rivaroxaban across VTE prophylaxis, AF stroke prevention, and DVT/PE treatment; rank candidates and assess comparative evidence.
2Key Finding
Scout screened 874 deduplicated trials across four direct Factor Xa inhibitors and produced a transparent, weighted evidence ranking.
Evidence sources: ClinicalTrials.gov (874 deduplicated trials); Open Targets; ChEMBL; Pharos; F10 target record (UniProt P00742).
Tool chain: F10 target ID → trial metadata retrieval → per-candidate summaries → weighted comparative ranking (maturity 50% · breadth 35% · adoption 15%).
Confidence: High for trial-evidence breadth and the apixaban ranking; Moderate for comparative safety/efficacy (few direct head-to-head outcomes; counts are metadata, not efficacy).
Still to confirm: Indication-specific comparative analysis using the identified COBRRA / COVET head-to-head studies.
3Evidence-Ranked Candidates
| Candidate | Trials | Composite | Recommendation |
|---|---|---|---|
| Apixaban | 406 | 82.15 | LEAD all indications covered |
| Edoxaban | 165 | 59.23 | STRONG #2 AF/VTE, pivotal warfarin data |
| Rivaroxaban | 553 | 86.76 | Reference agent |
| Betrixaban | 14 | 32.93 | Niche; smaller evidence base |
4Why Apixaban Leads
| Trial | Registry | Result |
|---|---|---|
| ARISTOTLE | NCT00412984 | Superior to warfarin for stroke/SE, less bleeding, lower mortality |
| ADVANCE-3 | NCT00423319 | Ortho VTE prophylaxis vs enoxaparin (9.0% vs 8.8%; HR 1.02) |
| CARAVAGGIO | NCT03045406 | Cancer-associated VTE (1,170 patients) |
| COBRRA / COVET | NCT03266783 / NCT03196349 | Direct apixaban-vs-rivaroxaban comparisons available |
5Recommended Path
- Lead with apixaban for a development or clinical-substitution program.
- Edoxaban as the AF/VTE alternative (ENGAGE AF-TIMI 48, Hokusai-VTE).
- Next step: indication-specific comparison using the identified head-to-head studies.
Not a chatbot. An agentic operating system.
Scientist-in-the-Loop
Scout augments expert judgment, never bypasses it. ASK_USER pauses when the planner encounters genuine ambiguity. The scientist directs; the AI executes.
Hundreds of Specialized Agents
Not one model trying to do everything. Hundreds of purpose-built computational agents — ADMET, docking, retrosynthesis, expression profiling, literature mining — orchestrated by an LLM planner, with more added every week.
Two Modes of Operation
Hand-craft validated pipelines that run autonomously across hundreds of compounds. Or let Scout plan from a research question. The pipeline handles volume; the planner handles novelty.
See Scout answer your research questions
Schedule a live demo with a real research question. No slides. Just science.
Request a Demo