Statistical Inference of Mixture Models of Amino acid Conformations in Proteins

Proteins are biomolecules made of amino acid chains that fold spontaneously into complex 3D structures. A protein’s shape depends on the conformations of its amino acids. Modelling these conformations across experimentally observed proteins has many benefits, but current methods have several limitations. This thesis addresses this statistical modelling challenge using the Bayesian criterion of Minimum Message Length that combines inductive inference with information theory and lossless data compression. It develops an unsupervised MML framework that trades off model complexity with fidelity to explain the observed distributions of amino acid conformations.