Project description: Having the potential to predict drug metabolism accurately would not only reduce drug attrition, but also support the design of optimal drug molecules. In this project, you will focus on building machine learning models on Cytochrome P450s, one of the most common enzyme families, involved in drug metabolism. Publicly available datasets will be used for this purpose (https://tdcommons.ai/). You will be using molecular fingerprints and other physicochemical properties as descriptors in your models. These can be easily calculated by using the rdkit module in python. You will have the opportunity to explore different classical machine learning algorithms (e.g., Random Forests, Support Vector Machines, Neural Networks). As a part of your modeling tasks, you could also do feature importance analyses, which would help you gain insights regarding the chemical features, influencing metabolism. Though the primary focus would be on classical approaches, you could investigate the application of deep learning (e.g. Graph Convolutional Neural Networks) to build a multi-task model (subject to time and interests).
What you will learn: Working with Chemistry datasets, understanding molecular representations, acquiring knowledge on the background of drug discovery process; note: although a background in chemistry is not required, an interest in chemistry and pharmaceuticals is desired!
Timeline: c.a. 3 months (one term, from late August to November)
Contact: industry mentors (AstraZeneca):
- Dr. Vigneshwari Subramanian, Senior Data Scientist, Data Science and AI group (Clinical Pharmacology and Safety Sciences), Sweden.
- Dr. Avid Afzal, Senior Data Scientist, Data Sciences and Quantitative Biology,Cambridge.