Doctor of Philosophy (PhD)
Electrical and Computer Engineering
Fast and accurate classification of ligand-binding sites in proteins with respect to the class of binding molecules is invaluable not only to the automatic functional annotation of large datasets of protein structures but also to projects in protein evolution, protein engineering and drug development. However, the complex structures of the ligand-binding sites and uncountable possibilities of chemical reactions makes developing a computational ligand-binding site classifier a challenging task. Another more challenging task is to efficiently predict or select binding drugs given a target ligand-binding sites. Artificial intelligence has already been successfully applied to address challenging problems across various fields where the the data structures are heterogeneous and complex. They are inherently suitable to provide solutions to these problems. Deep learning, a technique to train deep neural networks, is the main technology that drives concurrent artificial intelligence. In this research, three works based on state-of-the-art deep learning technologies are proposed. The first two provide solutions for ligand-binding sites classification, and the third describes a method to predict binding drugs for target-based drug design. The first approach develops image representations of ligand-binding sites and uses a convolutional neural network as the classifier. The second approach is based on graph; the ligand-binding sites are represented as non-Euclidean graphs, and the classifier is a customized graph neural network. In addition to classification applications, the third work describes a generative model to directly predict binding drugs for ligand-binding sites. All the methods exhibit a significant level of novelty, and are evaluated properly and thoroughly.
Shi, Wentao, "Artificial Intelligence for the Target-based Drug Design" (2022). LSU Doctoral Dissertations. 5794.
Available for download on Wednesday, March 28, 2029