Biological Signal Detection for Protein Function Prediction Investigators: Yang Dai Prime Grant Support: NSF Problem Statement and Motivation Technical Approach Key Achievements and Future Goals High-throughput experiments generate new protein sequences with unknown function prediction In silico protein function prediction is in need Protein subcellular localization is a key element in understanding function Such a prediction can be made based on protein sequences with machine learners Feature extraction and scalability of learner are keys. Use Fast Fourier Transform to capture long range correlation in protein sequence Design a class of new kernels to capture subtle similarity between sequences Use domains and motifs of proteins as coding vectors Use multi-classification system based on deterministic machine learning approach, such as support vector machine Use Bayesian probabilistic model Developed highly sophisticated sequence coding methods Developed an integrated multi-classification system for protein subcellular localization Developed a preliminary multi-classification system for subnuclear localization Will incorporate various knowledge from other databases into the current framework Will design an integrative system for protein function prediction based on information of protein localizations, gene expression, and protein-protein interactions Sequences specific subcellular and subnuclear localization MASVQLY... …HKEPGV Machine Learner Text File of Protein description Coding Vector s

Computational Protein Topographics for Health Improvement Jie Liang, Ph.D. Bioengineering Prime Grant Support: National Science Foundation Career Award, National Institutes of Health R01, Office of Naval Research, and the Whitaker Foundation. Problem Statement and Motivation Key Achievements and Future Goals The structure of proteins provide rich information about how cells work. With the success of structural genomics, soon we will have all human proteins mapped to structures. However, we need to develop computational tools to extract information from these structures to understand how cell works and how new diseases can be treated. Therefore, the development of computational tools for surface matching and for function prediction will open the door for many new development for health improvement. We have developed a web server CASTP (cast.engr. uic.edu) that identify and measures protein surfaces. It has been used by thousands of scientists world wide. We have built a protein surface library for >10,000 proteins, and have developed models to characterize cross reactivities of enzymes. We also developed methods for designing phage library for discovery of peptide drugs. We have developed methods for predicting structures of beta-barrel membrane proteins. Future: Understand how protein fold and assemble, and designing method for engineering better proteins and drugs. Technical Approach We use geometric models and fast algorithm to characterize surface properties of over thirty protein structures. We develop evolutionary models to understand how proteins overall evolve to acquire different functions using different combination of surface textures. Efficient search methods and statistical models allow us to identify very similar surfaces on totally different proteins Probablistc models and sampling techniques help us to understand how protein works to perform their functions. Evolution of function Protein surface matching

Structural Bioinformatics Study of Protein Interaction Network Investigators: Hui Lu, Bioengineering Prime Grant Support: NIH, DOL Problem Statement and Motivation Technical ApproachKey Achievements and Future Goals Protein interacts with other biomolecules to perform a function: DNA/RNA, ligands, drugs, membranes, and other proteins. A high accuracy prediction of the protein interaction network will provide a global understanding of gene regulation, protein function annotation, and the signaling process. The understanding and computation of protein-ligand binding have direct impact on drug design. Data mining protein structures Molecular Dynamics and Monte Carlo simulations Machine learning Phylogenetic analysis of interaction networks Gene expression data analysis using clustering Binding affinity calculation using statistical physics Developed the DNA binding protein and binding site prediction protocols that have the best accuracy available. Developed transcription factor binding site prediction. Developed the only protocol that predicts the protein membrane binding behavior. Will work on drug design based on structural binding. Will work on the signaling protein binding mechanism. Will build complete protein-DNA interaction prediction package and a Web server. Protein-DNA complex: gene regulation DNA repair cancer treatment drug design gene therapy