NODAGS-Flow: Nonlinear Cyclic Causal Structure Learning
This addresses the challenge of modeling real-world systems with feedback loops, such as in biology, by enabling nonlinear cyclic causal learning, which is an incremental advance over existing linear or acyclic methods.
The paper tackles the problem of learning causal relationships in systems with feedback loops by proposing NODAGS-Flow, a framework for nonlinear cyclic causal structure learning from interventional data, showing significant performance improvements in structure recovery and predictive performance compared to state-of-the-art methods.
Learning causal relationships between variables is a well-studied problem in statistics, with many important applications in science. However, modeling real-world systems remain challenging, as most existing algorithms assume that the underlying causal graph is acyclic. While this is a convenient framework for developing theoretical developments about causal reasoning and inference, the underlying modeling assumption is likely to be violated in real systems, because feedback loops are common (e.g., in biological systems). Although a few methods search for cyclic causal models, they usually rely on some form of linearity, which is also limiting, or lack a clear underlying probabilistic model. In this work, we propose a novel framework for learning nonlinear cyclic causal graphical models from interventional data, called NODAGS-Flow. We perform inference via direct likelihood optimization, employing techniques from residual normalizing flows for likelihood estimation. Through synthetic experiments and an application to single-cell high-content perturbation screening data, we show significant performance improvements with our approach compared to state-of-the-art methods with respect to structure recovery and predictive performance.