Persistent homology and machine learning for structure-based biomolecule property predictions
For more information about this meeting, contact Chuanbin Li.
Speaker: Zixuan Cang, Michigan State University
Abstract: The computational prediciton of biomolecule properties is important in many applications such as protein design and drug discovery. Machine learning, especially deep learning have shown their power in finding the map from descriptions to target quantities for different objects, given sufficient data. With the fast growing biomolecular databases, we apply machine learning methods for the structure-based prediction of biomolecule properties. We use persistent homology as a translator to provide an appropriate (concise and informative) description of the biomolecules to the machine learning algorithms. This topological method reduces the geometric complexity and addresses the properties of the interaction networks, which delivers features with suitable sizes while retaining the biological information. In this talk, I will describe the workflow of this approach and some applications, mutation induced protein stability change prediction, protein-ligand binding affinity prediction, and structure-based virtual screening.
Room Reservation Information
Room Number: 023
Time: 4:00pm - 5:00pm