[Research seminar today, Zoom, 4:00-5:00 PM] Biologically explainable dynamical systems underlying gene regulation in cancer.

[Wei Ding]

Title:  A biologically explainable formulation and sparsification of dynamical systems underlying gene regulation in cancer.

Zoom: https://umassboston.zoom.us/j/96508522662 

Time: Monday, March 4th, 2024, from 4:00 to 5:00 PM EST

Talk abstract:

This talk aims to answer the question “How can we build better machine learning models to describe the evolution of gene expression over time?”. Such dynamics models, which can be described by ordinary differential equations (ODEs), allow a deeper understanding of disease progression and response to therapy, thus aiding in intervention optimization. Although there exist many methods to infer these ODEs, these are generally limited to small networks of genes, rely on dimensional reduction, or impose non-biological parametric restrictions — all impeding scalability and explainability. We construct PHOENIX, a neural ODE framework incorporating prior domain knowledge as soft constraints to infer sparse, biologically explainable dynamics. We define explainability mathematically, and then demonstrate how our framework heavily outperforms currently used tools in this space. We then investigate approaches for inducing sparsity in this model, borrowing heavily from the pruning/lottery ticket literature. We develop a novel approach called DASH that can leverage domain-specific structural information to discover biologically meaningful lottery tickets. Finally, we demonstrate how PHOENIX+DASH can scale up to the entire human genome to uncover interpretable dynamics in breast cancer that are meaningful in the context of biological pathways.

Short bio:

I completed my undergraduate degree from Williams College (Massachusetts, USA) in 2017, with majors in Mathematics and Biology. My senior thesis at Williams was in Computational Biology where I performed a comparative genomics analysis of codon influence biases across multiple organisms. After undergrad, I spent two years in Analysis Group Inc (Boston), where I collaborated with Biostatisticians to develop prognostic models for rare diseases and also created software for streamlined statistical analysis of clinical data. Subsequently in 2019, I started a Biostatistics PhD at Harvard University, where I joined the network biology lab, under the supervision of John Quackenbush. Since then, I have worked on developing scalable and explainable machine learning tools for the discovery of dynamical systems underlying important biological processes such as cancer progression. I have also worked on connecting these dynamical systems to the realm of neural network pruning and sparse lottery tickets. On the side, I also enjoy learning about other areas of math, and have worked on applied projects in parallel computing and mixed-integer optimization. 

Best Regards, 

Wei Ding, Fellow of IEEE

Distinguished Professor, Computer Science

Executive Director, Paul English Applied AI Institute

University of Massachusetts Boston

100 Morrissey Blvd.

Boston, MA 02125-3393

Phone: 617.287.6428 

Webpage: www.cs.umb.edu/~ding