Project V: Probabilistic and Statistical Aspects of Protein Structure Prediction

Protein structure prediction utilizes a number of statistical and computational techniques: Hidden Markov Models (HMMs), probabilistic learning, neural networks, classification and model-based clustering. When a wide range of information sources is available, care should be taken to ensure that inference treats and combines the different sources of information appropriately. Perhaps more importantly, the nature of the statistical/computational learning approach used should be tailored specifically toward the learning goal, that is, the use of training samples and/or expert opinion concerning structural classification should be specifically oriented towards the inference problem being solved. For example, when training a HMM on sequences of known higher-order structure, the choice of training samples should be selected for optimal discriminatory/classification power.

This project will study probabilistic learning approaches to protein structure prediction and analysis, with reference to analysis packages and servers currently available, with the goal of finding an optimal learning strategy.