Generative Intervention Models for Causal Perturbation Modeling

2024-11-21

arXiv.org

AI Summary

This work addresses causal prediction under external interventions with unknown mechanisms but known intervention features (e.g., drug properties are known, while targeted biological pathways are unknown). Method: We propose the Generative Intervention Model (GIM), which directly maps intervention features to atomic intervention distributions and jointly learns both causal structure and intervention effects—without assuming predefined intervention types or prior knowledge of the causal graph. GIM integrates generative modeling with joint causal structure learning. Contribution/Results: The model achieves strong out-of-distribution (OOD) generalization to unseen interventions while retaining mechanistic interpretability. On synthetic benchmarks and single-cell RNA-seq drug perturbation datasets, GIM matches black-box models in OOD predictive performance and significantly improves accuracy in inferring true intervention mechanisms. To our knowledge, GIM is the first end-to-end, prior-free, and interpretable generative framework for causal perturbation modeling.

Technology Category

• Machine Learning: Causal Learning
• Reasoning under Uncertainty: Causality
• Knowledge Representation and Reasoning: Action, Change, and Causality

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• Predicting perturbation effects via causal models
• Mapping perturbation features to intervention distributions
• Inferring underlying perturbation mechanisms from data

Innovation

• Generative model maps perturbation features to interventions
• Jointly estimates causal model and perturbation effects
• Robust out-of-distribution predictions with mechanistic insights