Arman Bilge



Broadly, I am interested in developing computational methods to enhance our understanding of evolutionary processes. In particular, I am eager to advance techniques for Bayesian phylogenetic inference.

Cophylogeny Reconstruction

Cophylogenies are systems of interdependent evolutionary trees and are relevant to models for coevolution between symbiotic organisms and genome evolution. Specifically, a complete cophylogenetic history consists of a mapping of one evolutionary tree onto another and the sequence of biological events that generated this mapping (e.g., symbiotic host-switches or horizontal gene transfers).

Previous methods for cophylogenetic reconstruction generally attempt to find the simplest explanation. However, developing a probabilistic approach to this problem would enable reconstruction methods to consider more complex explanations as well as provide an idea of the statistical uncertainty. The challenge is to find an efficient way to average over the infinite number of histories including unobserved events that result in the same observed history.

Hamiltonian Monte Carlo

Bayesian phylogenetic inference involves sampling parameters for the evolutionary model according to their probability given the data. This sampling is usually done with the Markov chain Monte Carlo (MCMC) algorithm. We can imagine a frictionless skatepark where each point in the park corresponds to a particular set of parameter values for our model. Specifically, we will design it such that parameter values with high probability map to points in the park with low elevation. Then the MCMC algorithm looks like a random walk that moves through the park very slowly, thus making it an inefficient sampler.

Instead, we can place a skater in this park whose movement is given by Newton’s laws of motion. They will traverse the area efficiently by taking advantage of the features of the skate park. The Hamiltonian Monte Carlo (HMC) algorithm is a sampler that uses this technique. In my implementation of HMC for phylogenetics, it offers on average five times better performance than MCMC.

I presented a poster on “Efficient Bayesian Evolutionary Analysis using Hamiltonian Monte Carlo” as a part of the symposium on Untangling information, noise, and phylogenetic reconstruction in genome scale data at SMBE 2015 in Vienna.

My work was partially supported by a Summer Scholarship from The Allan Wilson Centre for Molecular Ecology and Evolution.