Advancing Machine Learning Research for Biological Sciences: We develop next-generation machine learning tools tailored for biological research. Our focus is on improving causality, interpretability, disentanglement, uncertainty quantification, and decision-making in machine learning models to enhance their robustness and scientific utility.
Designing Innovative Computational Tools for Molecular Biology Research: We leverage cutting-edge genetic engineering and high-throughput profiling technologies, such as CRISPR and single-cell RNA sequencing, to study complex diseases and drive drug discovery. By integrating computational methods with experimental biology, particularly in immunology, we aim to make significant advances in understanding cellular processes and disease mechanisms.
We strive to address these challenges in our main areas of research laid out below.
We develop ML methodology for making generative models more interpretable and usable for downstream tasks such as decision-making and hypothesis testing. These models are particularly useful in handling high-dimensional, noisy, and incomplete data typical in applied scientific research.
We develop causal machine learning approaches that can leverage high-dimensional data. Towards this goal, we are interested in tractable approaches to causal structure learning that have the potential to scale to tens of thousands of variables. Additionally, we are interested in causal representation learning, where interventions are conducted on latent variables of a deep generative model.
Our lab develops advanced algorithms to analyze single-cell omics data, enhancing our understanding of cellular states and dynamics. We focus on improving methods for differential expression analysis, integration of multi-omics data, and robust modeling of cellular heterogeneity. These innovations are vital for deciphering the complexities of single-cell data and driving biological discoveries.
Leveraging spatial transcriptomics, we aim to map cellular organization within tissues, combining computational biology techniques with experimental data to uncover spatial patterns and interactions at the molecular level. Our research focuses on developing robust methods for analyzing spatially resolved transcriptomics data, leading to new insights into tissue architecture and cellular function.
We explore the effects of genetic and chemical perturbations at the single-cell level, developing models that can predict cellular responses to these perturbations. This research helps in understanding the mechanisms of action for various perturbations, aiding in drug discovery and therapeutic interventions. Our models aim to be robust, interpretable, and applicable across different biological contexts.