Charles McAnany, PhD
Charles McAnany, PhD
Postdoctoral Research Associate
Stowers Institute for Medical Research
Charles McAnany holds a Ph.D. in Chemistry from the University of Virginia, where he specialized in molecular dynamics simulations of biomolecular assemblies and intrinsically disordered proteins. Since 2019, he has been working with Dr. Julia Zeitlinger at the Stowers Institute for Medical Research to apply machine learning techniques to genomic data in order to understand the sequence rules underlying gene regulation. His research touches on developmental signaling, structural biophysics, high-performance computing, cellular adhesion, ribonucleoprotein assembly evolution, and automated bias removal in experimental data. In his free time, he is an avid photographer, with a full darkroom in his basement capable of processing and printing negatives from his 4×5 camera.
Visualizing the Cis-Regulatory Code
Deep learning models have been successfully applied to genome-wide data, revealing interactions between transcription factors, rules of promoter behavior, nucleosome positioning sequences, and many other regulatory patterns. A key challenge in using deep learning in genomics is the difficulty of interpretation: Even if a model makes high-accuracy predictions, the internal representations it uses are highly distributed and difficult to understand directly. I have developed a technique visualize the learned sequence rules in two dimensions, showing the effect of each base in the DNA on every (predicted) property at its neighbors. These pairwise interaction maps provide a direct and visual readout of the sequence rules underpinning the cis-regulatory code, and can be used to view complex, long-range effects of multiple sequence elements. For experimental biases with short-range effects, these maps also serve to separate those biases from the underlying biological processes, even in cases where the precise nature of the bias is not known a priori. These bias-reduction techniques can be used to predict underlying biological phenomena that are otherwise hidden by experimental artifacts.