Multi-label Bioactivity Prediction from Multi-site Cell Painting Data via Semi-Supervised Contrastive Learning and Ensemble Learning

This repo contains the code to reproduce results from our paper Multi-label Bioactivity Prediction from Multi-site Cell Painting Data via Semi-Supervised Contrastive Learning and Ensemble Learning. In this work, we demonstrate the application of MuSWSemiSupCon models for extracting meaningful features from Cell Painting image data, to facilitate accurate bioactivity prediction. MuSWSupCon extends our previously published SemiSupCon framework. The implementation of SemiSupCon is publicly available on GitHub at [https://github.com/AGSun-FMP/CP_SemiSupCon].

Baseline comparisson on FMP Cell Painting dataset

Data

Images

EU-OPENSCREEN Bioactive Cell Painting dataset

EU-OPENSCREEN Bioactives HepG2 microscopy images processed with ImageJ

BBBC022 images processed with ImageJ

Embeddings

MuSWSupCon(EU-OS bioactives) Embeddings