Skip to content

Experimental design & power analysis for perturbation screens #1037

Description

@Zethson

Context

Before spending significant money on a Perturb-seq or drug screen, users need to plan it, and afterwards justify that it was adequately powered.
pertpy has no tooling for this today, and little exists anywhere — largely greenfield.

What's missing

  • A priori power / sample-size calculation: given an expected effect size (e.g. in E-distance / logFC terms) and variability, how many cells per perturbation and how many replicates are needed to detect it at a target power?
  • Post-hoc power / minimal detectable effect from a pilot dataset.
  • Perturbation selection / active learning: with a prediction model, which perturbations to include in the next round of an iterative screen to maximize information.

Proposal / API

A pertpy.tools design module.
Power calculations can be simulation-based, reusing Distance / DistanceTest to define effect sizes consistent with the rest of pertpy.
Active-learning selection builds on the prediction + perturbation-space APIs.

Why it matters

  • Industry: directly reduces wasted screening spend and supports go/no-go decisions.
  • Academia: reviewers increasingly ask for power justification, and iterative screen design is an active research area.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

enhancementNew feature or request

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions