Disentangling Covariates to Predict Counterfactuals for single-cell data
Keywords: single-cell, computational biology, causal inference, generative model, variational inference, variational autoencoder, fairness, representation disentanglement
TL;DR: dis2p disentangles known covariate variations from unknown ones while simultaneously learning to make counterfactual predictions.
Abstract: Single-cell transcriptomics enables understanding cellular behaviors during diseases or in response to perturbations. However, analyzing multi-donor and multi-covariate single-cell data while disentangling technical noise from biological signals remains a significant challenge. Additionally, predicting cellular responses to interventions becomes even more challenging due to donor-specific effects and unobserved covariates. This study introduces “disentanglement to prediction” (dis2p), a causal generative model designed to disentangle known covariate variations from unknown ones while simultaneously learning to make counterfactual predictions. dis2p accurately learns covariate-specific representations, as empirically demonstrated, which improve generalization for performing counterfactual predictions. Given the increasing availability of population-level single-cell datasets, we envision dis2p becoming a valuable tool for analyzing such data due to its ability to learn controllable representations that facilitate biological discoveries, improve experiment design, and reduce costs using in silico predictions.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 9140
Loading