Predicting functional sites in proteins using intelligent machine learning and data mining techniques

Proteins are fundamental building blocks and play a central role in defining behavior within all biological systems. Identification of functional residues is a crucial step towards understanding the functional mechanisms of proteins. Experimental elucidation of protein function can be prohibitively expensive. To address this, this thesis focuses on developing computational approaches to predict protein functional sites from sequence and/or structural information, by combining feature engineering and modern machine-learning algorithms. Specific techniques are developed for predicting functional sites of different types. These demonstrate the importance of leveraging heterogeneous features. Online servers have been created that make the developed computational techniques publicly available.