We aim to improve drug-candidate selection by predicting the effect of drug perturbations on a virtual cell based on single-cell sequencing data. If successful, this should enable the development of more and better drugs.

We are currently iterating on several models (autoencoders, drug embeddings, flow matching models) that work together to predict the effect of drug perturbations. Over the next months, we aim to scale up these approaches by several orders of magnitude both in data and compute.

We met as students of the computational biology masters @ ETH and bonded during long, shared hours of exam preparation. We both care deeply about improving the grossly inefficient drug discovery process.