Predictable perturbations

While technologies like Perturb-seq let scientists observe cells' responses to genetic changes at scale, the ability to predict those responses using machine learning tools could greatly accelerate research on disease mechanisms.

Former Schmidt Center MEng Emily Liu, current PhD fellow Jiaqi Zhang, and Director Caroline Uhler have developed a new model called Single Cell Causal Variational Autoencoder (SCCVAE) that addresses two key challenges in modeling unseen genetic perturbations: lack of generalizability, and difficulty with noisy, large-scale single cell data. By integrating deep learning and mechanistic approaches, SCCVAE allows the team to simulate new experiments and reveal groups of genes that work in concert.

Learn more in PLOS Computational Biology.

Adapted from an updated written by Broad Communications.