A foundation model for functional drug response
Trained on diverse functional screening experiments across cell types, assay modalities, and small molecule classes. Fine-tuned on your data. Deployed against your compound library.
Where phenotypic prediction fits
Complementary to structural and target-based tools. A distinct axis of biological information.
Most drugs fail because screening captures a fraction of biology
Single-assay workflows rarely capture the biology that drives lead failure: off-target liabilities, pathway effects, and context-dependent cellular response.
Our model predicts phenotypic response across the full landscape of cellular contexts, helping discovery teams make better lead-prioritization decisions.
Traditional screening
One protocol. One context.
Reflector
Every cell type. Every modality. Simultaneously.
From your data to prioritized candidates
Four steps. Four weeks.
Ingest
Your existing screening data: functional assays, dose-response curves, protocols.
Fine-tune
Our model adapts to your assay with as few as 50 small molecules and a quantitative endpoint.
Predict
Using chemical structure, we score your virtual library and rank compounds by predicted assay response. Each prediction includes a calibrated confidence estimate.
Deliver
Ranked candidates with predicted activity, flagged liabilities, and off-target annotations. Ready for experimental follow-up.
What the platform does
Virtual functional screening
Augment your primary screen with phenotypic predictions across your entire compound library. Our model ranks compounds by expected assay response, helping teams prioritize the best candidates for synthesis and follow-up screening.
We use chemical structure to score compounds that have never been synthesized, enabling virtual screens at scales not possible experimentally.
Compound risk profiling
Evaluate leads across diverse cellular contexts before committing to follow-up assays. Flag off-target signals, toxicity, and context-dependent effects earlier in discovery.
A compound that looks promising in one disease-relevant model may fail in hepatocyte, cardiomyocyte, or immune contexts. Our platform helps identify those risks earlier, so teams can prioritize stronger leads with fewer downstream surprises.
Mechanism identification
Each compound screen through our platform generates a unique activity-based fingerprint. That makes it possible to compare compounds by phenotypic behavior, not just by chemical structure.
Teams can use those fingerprints to identify new indications for existing assets, revisit shelved programs, and uncover mechanistic hypotheses across a portfolio without relying on target annotation.
How collaborations are scoped
Scoping call
30 minutes to assess your biology, available data, and what a useful outcome looks like. We determine fit before either side commits.
Prediction and delivery
We assess quality, adapt the model to your assay, and score your compound library. You receive ranked candidates with predicted activity and flagged liabilities, ready for experimental follow-up.
Iteration
Experimental results feed back into our model. Each validation cycle improves predictions for subsequent rounds.
What are you working on?
Tell us about your biology. We will tell you whether our model already covers it, or what it would take to get there.