Seedlabs

Context-Screen: Zero-Shot Cell-Context Prioritizer for Compound Perturbations

A report that predicts which types of cells a new drug will affect, helping labs pick the best models to test.

A screening report that predicts which cellular contexts a candidate compound will strongly perturb — including disease contexts where that compound has never been tested — so pharma teams can prioritize which cell models to run in the wet lab.

Biochemistry, Genetics and Molecular BiologySingle-cell and spatial transcriptomics
Cross-Corpus Speech Emotion RecognitionPharmaceutical lead profiling and perturbation-screen experimental design

Concept

A decision-support screening service for pharma and biotech perturbation programs. The customer supplies a compound — by identity or, more powerfully, by a perturbation signature measured in any one convenient context (e.g., a standard cell line). The service runs the State model to predict the strength and differentially-expressed-gene profile of that perturbation across a broad panel of cellular contexts: cell types, donors, and disease states. Because the underlying cell embeddings were learned from observational data, the screen works even for contexts where no perturbation data exists for that compound. Deliverable: a ranked shortlist of "hot" contexts where the compound is predicted to act strongly, with predicted DE-gene lists per context, so teams spend Perturb-seq and validation budgets only where the model says the biology will move. A secondary readout flags healthy contexts with strong predicted effects as potential off-target toxicity signals. The product is a prioritization layer that shrinks wet-lab scope — not a replacement for experiments.

Why now

Paper [0] demonstrates exactly the missing capability: State predicts perturbation effects while accounting for cellular heterogeneity within and across experiments, improved discrimination of effects on large datasets by more than 30% versus baselines, and identified differentially expressed genes across chemical (as well as genetic and signaling) perturbations with significantly improved accuracy. Most importantly for this product, its cell embeddings — trained on observational single-cell data from 167 million cells — enable identification of strong perturbations in cellular contexts where no perturbations were observed during training. Since the vast majority of compound–context pairs have never been measured, this zero-shot capability is what makes a context-prioritization screen feasible today rather than a research aspiration.

Written by GLM · 12 Oct 2026

AI assessment

Backed by 1 paper94

A high-value decision-support tool that leverages zero-shot cellular embeddings to reduce expensive wet-lab screening costs for pharma lead profiling.

Evidence strength
5/5
The idea directly implements the core capability of the State model, specifically its ability to predict perturbations in contexts where no prior perturbation data exists using observational embeddings.
Market pull
5/5
Pharma companies (e.g., Recursion, AstraZeneca) spend millions on Perturb-seq and high-throughput screening; a tool that narrows the search space has immediate ROI.
Novelty & moat
4/5
While other catalog ideas focus on specific cell types or toxicity, this is a general-purpose zero-shot prioritization engine for any compound-context pair.
Feasibility
4/5
The model (State) and evaluation framework (Cell-Eval) already exist, meaning the MVP is primarily a software wrapper around an existing architecture.
Wedge clarity
5/5
The wedge is a specific 'prioritization report' for wet-lab experimental design, avoiding the trap of claiming to replace the lab entirely.
Simplicity / focus
5/5
The product is a single, focused service: input a compound/signature, output a ranked list of high-probability cellular contexts.

Scored by Gemma 4 · 12 Oct 2026. Rated against a fixed rubric (evidence, market, novelty, feasibility, wedge, simplicity), relative to the closest ideas already in the catalog. A prior estimate to compare ideas before real-world signal arrives.

Persona discussion

AI personas trained on real people's expertise debate this idea as it evolves.

View the discussion →

Act on this idea

Ideas only matter if someone runs with them. Your message goes straight to the founder's inbox — nothing is stored on our servers.

Who benefits

  • Recursion Pharmaceuticalscompany

    Operates one of the industry's largest proprietary perturbation datasets and ML triage pipelines; a zero-shot context screen tells them which disease contexts each hit perturbs before they commit wet-lab spend.

  • insitrocompany

    Builds single-cell disease models to select targets and compounds; predicting strong perturbation effects in untested patient-derived contexts sharpens their validation choices.

  • AstraZenecacompany

    Runs large perturb-seq and CRISPR screening programs; a ranked context shortlist reduces screen scope and cost per program.

  • Broad Instituteorganization

    Maintains public Perturb-seq and DepMap resources; a context-prioritization tool helps design which cellular contexts to profile next for the community.

Research it builds on

  1. Predicting cellular responses to perturbation across diverse contexts with State
    Abhinav Adduri, Dhruv Gautam, Beatrice Bevilacqua et al. · 2026 · 89 citations
    All ideas from this paper →
Leave feedback
feasibility