The development of entirely new drugs is a costly and arduous process. A different strategy relies on combining already approved drugs in so-called combination therapies. Here, the challenge is identifying promising combinations for clinical testing.

To do that more efficiently, the lab of Fabian Theis, scientific director of Helmholtz AI, teamed up with Facebook AI to develop the Compositional Perturbation Autoencoder (CPA). Employing a self-supervised machine learning technique CPA encodes and learns transcriptional drug responses across different cell types, doses, and drug combinations. This new tool combines the interpretability of linear models with the flexibility of deep-learning approaches for single-cell response modelling. That knowledge can guide experimental testing and help to discover new treatment options for complex diseases.

Paper: https://www.biorxiv.org/content/10.1101/2021.04.14.439903v1

Dataset: https://github.com/facebookresearch/CPA

Blogpost at Facebook AI: https://ai.facebook.com/blog/ai-predicts-effective-drug-combinations-to-fight-complex-diseases-faster