Generative Intervention Models for Causal Perturbation Modeling

Nora Schneider, Lars Lorch, Niki Kilbertus, Bernhard Schölkopf, Andreas Krause

TL;DR: We introduce a novel framework for predicting general perturbation effects via causal models.

Abstract: We consider the problem of predicting perturbation effects via causal models. In many applications, it is a priori unknown which mechanisms of a system are modified by an external perturbation, even though the features of the perturbation are available. For example, in genomics, some properties of a drug may be known, but not their causal effects on the regulatory pathways of cells. We propose a generative intervention model (GIM) that learns to map these perturbation features to distributions over atomic interventions in a jointly-estimated causal model. Contrary to prior approaches, this enables us to predict the distribution shifts of unseen perturbation features while gaining insights about their mechanistic effects in the underlying data-generating process. On synthetic data and scRNA-seq drug perturbation data, GIMs achieve robust out-of-distribution predictions on par with unstructured approaches, while effectively inferring the underlying perturbation mechanisms, often better than other causal inference methods.

Lay Summary:

Understanding how systems respond to small changes is a key goal in many scientific fields — for example, predicting how a cell reacts to different drugs. Since testing every possible drug experimentally is infeasible, we need computational methods to make such predictions. Often, we have data showing how a system responds to a set of known drugs, along with features that describe these drugs, such as their chemical properties. The challenge is to predict how the system will change under a new, previously unseen drug.
We introduce an approach, called Generative Intervention Models (GIMs), that learns the system's internal structure and how different perturbations (the drugs) affect it. Specifically, we model the system using a causal graph that captures how variables, such as the genes of cells, influence each other. Then, using the features of each drug, we learn how that drug alters the system’s internal mechanisms.
This approach allows us to predict the effects of new drugs and gain insight into how they act on the system. On synthetic and single-cell RNA sequencing (scRNA-seq) drug data, GIMs show strong performance and accurately infer the underlying mechanisms of the perturbations.

Primary Area: General Machine Learning->Causality

Keywords: Causality, Causal Inference, Causal Modeling, Interventions, Perturbations, Generalization, Interpretability

Submission Number: 2470