Gaussian Process Density Estimation to Obtain Ramachandran Probability Distributions

ARROYUELO, AGUSTINA.; VILA, JORGE A.; MARTÍN OSVALDO A.

San Luis

Conferencia; PyData San Luis 2017; 2017

Protein dihedral degrees of freedom play a central role in simulation andstructural analysis of these biomolecules with φ and ψ angles as the main torsionals influencing protein 3D structure. The Ramachandran map displays the sterically allowed regions and forbidden ones for the torsional angle combination (φ, ψ). Therefore featuring enlightenment about the conformation that every possible aminoacid in a protein structure can present. Sampling of protein conformational space using a Bayesian machinery requires the inclusion of a prior for the torsional angles (φ, ψ). Taking these torsionals from a uniform distribution is a very poor prior for this task. In the present work we explore the implementation of a non parametric model, such as Gaussian Process to learn a Ramachandran Probability Distribution for its use in Bayesian Inference. In future work we will test our method comparing with results of simulations using a uniform torsional angle prior. Our implementation uses Python?s probabilistic programming module, PyMC3.