Extended Data Fig. 3: Integrated gradients analysis.
From: Pertpy: an end-to-end framework for perturbation analysis
(a) Top 15 genes identified per gene program as most important for predicting the corresponding perturbation (single gene or gene pair) using the MLPClassifier. Gene importance was determined using integrated gradients (Methods). Attribution scores are shown for each gene, averaged across all cells within the respective gene program. Genes directly targeted by at least one perturbation within the gene program group are highlighted in red. (b) Pairwise MSE distances between gene programs in the perturbation space.
