SHERLOCK: Structured representation learning and causal inference of downstream perturbation effects

The article discusses a novel computational approach called SHERLOCK, which aims to provide a comprehensive framework for understanding the effects of various genetic, molecular, and experimental perturbations on cellular mechanisms. Traditional computational methods have focused on individual tasks, such as predicting perturbation responses, analyzing downstream transcriptional effects, or modeling perturbation combinations, without offering a unified approach for learning interpretable perturbation representations and enabling causal analysis of downstream effects.

SHERLOCK is a deep generative framework designed specifically for single-cell perturbation analysis. It represents perturbation effects as structured interventions on a latent baseline cellular state, allowing for the organization of genetic and pharmacological perturbations based on shared transcriptional responses. By formulating perturbations as interventions within a structural causal model, SHERLOCK enables counterfactual estimation of their downstream transcriptional effects under explicit identifiability assumptions.

Additionally, the same framework quantifies how perturbation responses vary across conditions and can compositionally model combinatorial perturbations, facilitating predictions of held-out combinations and classification of genetic interactions.

The effectiveness of SHERLOCK is demonstrated across multiple genome-scale CRISPR, chemical, and spatial perturbation datasets. It successfully recovers perturbation relationships that align with known biological pathways and pharmacological properties. The framework also identifies condition-dependent responses and accurately predicts the effects of combinatorial perturbations.

In summary, SHERLOCK offers a unified and interpretable approach for causal analysis of perturbation effects, making it a valuable tool for researchers working with single-cell perturbation experiments.