Teaming up with the machine: Using Domain Knowledge to Improve Predictions of Molecular Interactions Tomas Rube, UC Merced

Cells rely on millions of highly specific molecular interactions to process information and regulate biological processes. While new experiments can profile such interactions with high throughput, interpreting the resulting data is a major computational challenge. In this seminar, I will describe a flexible machine-learning framework that uses such data to define molecular recognition in terms of biophysically rigorous equilibrium binding constants or kinetic rates. Specifically, I will discuss how traditional mathematical modeling based on domain knowledge can be combined with methods from machine learning, and how the resulting hybrid framework makes it meaningful to perform previously uninterpretable experiments.