Code Repository for "A Pooled Cell Painting CRISPR Screening Platform Enables de novo Inference of Gene Function by Self-supervised Deep Learning"

Installation

Create environment, install dependencies and activate the environment

curl -fsSL https://pixi.sh/install.sh | bash
pixi install

# activating the environment for running shell scripts
pixi shell

# for running jupyter notebooks
pixi run jupyter lab

Dataset Availability

The datasets are publicly available in Amazon S3 at the following URLs:

• 124 gene CRISPR KO Morphology dataset
• 300 gene CRISPR KO Mechanism of Action (MoA) dataset
• 1640 gene CRISPR KO Druggable Genome dataset

POSH Barcode Sequencing and Assignment to CellPainting images

Tutorial for using the POSH barcode sequencing data to assign barcodes to CellPainting images can be found here

CP-DINO model

Scripts for Training CP-DINO model can be found here

Scripts for Inference using CP-DINO model can be found here

Analysis of CP-DINO Embeddings

Notebooks for analysis and evaluation of CP-DINO embeddings are available for the following datasets:

• Morphology 124 Dataset

  • Analysis notebook

• MoA 300 Dataset

  • Analysis notebook

• Druggable Genome 1640 Dataset

  • Analysis notebook

Training

Command for training CP-DINO model on 124 gene CRISPR KO Morphology dataset

Note: User may need to enable access to .sh scripts via chmod 700 ./scripts/training/cpdino*

./scripts/training/cpdino_300_fp32_100ep.sh
./scripts/training/cpdino_1640_fp32_100ep.sh
./scripts/training/cpdino_1640_with_pS6_fp32_100ep.sh

Pretrained Model Weights

Pretrained model weights for CP-DINO can be found here.

Inference

Command for inference using CP-DINO model on 124 gene CRISPR KO Morphology dataset

./scripts/inference/cpposh_124_inference.sh
./scripts/inference/cpposh_300_inference.sh
./scripts/inference/cpposh_1640_inference.sh