Data Science

Xuezhou Zhang

Assistant Professor of Computing & Data Sciences

Contact me

Options

Potential for UROP Funding

Overview

Background: Most orally dosed drugs conform to a narrow set of molecular properties. However, the past several years have seen growing interest in molecule types that deviate
from these conventional notions of “druglikeness”, as a means to address particularly challenging drug targets (mostly human proteins) of high therapeutic potential. Of particular interest are macrocyclic compounds (MCs) – molecules that contain a ring of size >12 atoms as part of their molecular structure. There is strong evidence that certain macrocyclic compounds possess favorable properties that enable them to be dosed orally and achieve systemic distribution despite violating conventional notions of “druglikeness”. The molecular origins of these advantageous properties are the subject of much research, but remain poorly understood. Consequently, there is a pressing need for informed guidance as to what kinds of MCs might be useful for drug discovery, and thus what kinds of macrocycles drug discoverers should be making and testing.

The scope of this undergraduate research project include: (1) develop novel interpretable machine learning algorithms to predict the membrane permeability of molecules based on existing manually extracted features and (2), in the case that the first stage goes well, discover new topological features that are predictive of molecule properties using recent Graph Neural Network (GNN) techniques.