Causal graphs provide a promising framework to understanding biological systems, especially in building more interpretable ML models when applied to hard biological tasks like understanding the mechanism through which a drug or therapy works. However, biological networks are extremely vast and complicated, meaning that the amount of quality data to develop such causal graphs is almost never available. Therefore, the incorporation of prior biological knowledge is likely needed to efficiently build such causal graphs. Using data driven approaches on a variety of dataset such as Perturb-Seq combined with prior knowledge such as known protein-protein interactions, I will develop scalable algorithms that learn meaningful causal graphs of gene regulatory networks (GRNs) despite noisy, limited data and incomplete, noisy prior knowledge. Then, I will benchmark these learned GRNs on downstream tasks such as gene perturbational outcome to determine if such networks are biologically meaningful and can be incorporated into biology-inspired machine learning.

Through this SuperUROP, I plan to learn more about causal inference – an area of research that is new to me but very promising as it has applications to more interpretable and causality-based methods in applying machine learning to biology. This project will inform and provide inspiration for my future research and career as I hope to develop causal-based AI / ML methods that are able to incorporate known biology for more interpretable and powerful models.