What Makes a Representation Good for Single-Cell Perturbation Prediction?

Wenkang Jiang, Yuhang Liu, Yichao Cai, Erdun Gao, Jiayi Dong, Ehsan Abbasnejad, Lina Yao, Javen Qinfeng Shi
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:52722-52757, 2026.

Abstract

Single-cell perturbation modeling is fundamental for understanding and predicting cellular responses to genetic perturbations. However, existing approaches, from causal representation learning to foundation models, often struggle with an overlooked challenge: gene expression is dominated by perturbation-invariant information, while perturbation-specific signals are intrinsically sparse. As a result, learned representations either entangle invariant and perturbation-specific information, leading to spurious and non-generalizable predictors, or suppress perturbation-specific signals altogether, rendering them ineffective for prediction. To address this, we propose PerturbedVAE, a general framework designed to resolve this signal imbalance. The framework explicitly separates perturbation-specific information from dominant invariant structure and recovers causal representations to effectively utilize such information for prediction. We further provide an identifiability analysis that characterizes the conditions under which sparse perturbation effects can be reliably recovered, thereby clarifying how the framework can be concretely specified under such conditions. Empirically, PerturbedVAE achieves state-of-the-art performance on a widely used benchmark across multiple evaluation settings, yielding significant gains on out-of-distribution combinatorial predictions and uncovering interpretable perturbation-response programs.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-jiang26ad, title = {What Makes a Representation Good for Single-Cell Perturbation Prediction?}, author = {Jiang, Wenkang and Liu, Yuhang and Cai, Yichao and Gao, Erdun and Dong, Jiayi and Abbasnejad, Ehsan and Yao, Lina and Shi, Javen Qinfeng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {52722--52757}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/jiang26ad/jiang26ad.pdf}, url = {https://proceedings.mlr.press/v306/jiang26ad.html}, abstract = {Single-cell perturbation modeling is fundamental for understanding and predicting cellular responses to genetic perturbations. However, existing approaches, from causal representation learning to foundation models, often struggle with an overlooked challenge: gene expression is dominated by perturbation-invariant information, while perturbation-specific signals are intrinsically sparse. As a result, learned representations either entangle invariant and perturbation-specific information, leading to spurious and non-generalizable predictors, or suppress perturbation-specific signals altogether, rendering them ineffective for prediction. To address this, we propose PerturbedVAE, a general framework designed to resolve this signal imbalance. The framework explicitly separates perturbation-specific information from dominant invariant structure and recovers causal representations to effectively utilize such information for prediction. We further provide an identifiability analysis that characterizes the conditions under which sparse perturbation effects can be reliably recovered, thereby clarifying how the framework can be concretely specified under such conditions. Empirically, PerturbedVAE achieves state-of-the-art performance on a widely used benchmark across multiple evaluation settings, yielding significant gains on out-of-distribution combinatorial predictions and uncovering interpretable perturbation-response programs.} }
Endnote
%0 Conference Paper %T What Makes a Representation Good for Single-Cell Perturbation Prediction? %A Wenkang Jiang %A Yuhang Liu %A Yichao Cai %A Erdun Gao %A Jiayi Dong %A Ehsan Abbasnejad %A Lina Yao %A Javen Qinfeng Shi %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-jiang26ad %I PMLR %P 52722--52757 %U https://proceedings.mlr.press/v306/jiang26ad.html %V 306 %X Single-cell perturbation modeling is fundamental for understanding and predicting cellular responses to genetic perturbations. However, existing approaches, from causal representation learning to foundation models, often struggle with an overlooked challenge: gene expression is dominated by perturbation-invariant information, while perturbation-specific signals are intrinsically sparse. As a result, learned representations either entangle invariant and perturbation-specific information, leading to spurious and non-generalizable predictors, or suppress perturbation-specific signals altogether, rendering them ineffective for prediction. To address this, we propose PerturbedVAE, a general framework designed to resolve this signal imbalance. The framework explicitly separates perturbation-specific information from dominant invariant structure and recovers causal representations to effectively utilize such information for prediction. We further provide an identifiability analysis that characterizes the conditions under which sparse perturbation effects can be reliably recovered, thereby clarifying how the framework can be concretely specified under such conditions. Empirically, PerturbedVAE achieves state-of-the-art performance on a widely used benchmark across multiple evaluation settings, yielding significant gains on out-of-distribution combinatorial predictions and uncovering interpretable perturbation-response programs.
APA
Jiang, W., Liu, Y., Cai, Y., Gao, E., Dong, J., Abbasnejad, E., Yao, L. & Shi, J.Q.. (2026). What Makes a Representation Good for Single-Cell Perturbation Prediction?. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:52722-52757 Available from https://proceedings.mlr.press/v306/jiang26ad.html.

Related Material