Zaikang Lin

Accurate prediction of cellular response to genetic perturbation has great implication in molecular biology and medicine. Modern high-throughput biological datasets with thousands of perturbations and multiple time points provide us with the ability to train deep-learning models that generalize to predict effects of unseen perturbations. Previous methods such as graph neural networks and low-rank neural ordinary differential equations had varying levels of success. But the overall predicative performances of models in such task remain low and some cannot train time-series data. I aim to develop a new archiecture that integrates the strengths of previous models, while making more accurate predictions on unseen perturbations in continuous time.