Statistical and computational methods for Bayesian phylogenetic inference - Belgium
- This project focuses on new developments ina popular Bayesian phylogenetic inference framework and its applications to important evolutionary problems, with a particular focus on infectious diseases.
- The first part of the project involves the development of an integrated web system and database that allows users to register, upload and retrieve sequence data to and from the database. Each user will be able to determine the sharing policy for the data he/she has provided.
- Such a system should be easily distributed so that other research groups can deploy it on their own server(s)and put it to use without technical interventions. Further, this system will interface with the BEAST software package to analyze the sequence data in inefficient manner.
- Additionally, visualization packages typically associated with BEAST, such as FigTree may be incorporatedinto this system.
- The candidate is expected to design and implement such system and determine an appropriate strategy to properly distribute the developed system as an easily installed/deployed software package.
- In the second part of the project, multipleparallelization ideas will be implemented in the BEAST software package.
- BEASTis mostly written in Java, with its high-performance computational library,known as BEAGLE, being implemented in C/C++ with specific routines that allow for parallel computing on both multi-core CPU and GPU platforms.
- Recent developments in thefield of phylogenetics have shown that straightforward palatalization approachesdo not make sufficient use of current multi-core architectures, both in CPU andGPU applications.
- The goal of this project is to tackle these issues from botha computational and a statistical perspective.
- The computational aspect entailsthe implementation of popular routines typically used in computer architecture,to perform automated load balancing and concurrent evaluation of possiblefuture states in the evaluation of posterior distributions.
- The usefulness of various look-ahead strategies will be implemented and evaluated against the current state of the art.
- The statistical aspect on the other hand entails the development and adaptation of novel transition kernels in a Bayesianphylogenetic inference framework.
- Along with the obtained computational improvements, better-performing transition kernels may aid in reducing run times of Bayesian analyses by exploring (posterior) distributions of interest in a more efficient manner.
- The candidate will be able to perform research in a dynamic and multidisciplinary team (computer scientists, data analysts and evolutionary biologists), housed in the brand new facilities of the Rega institute at the University hospital campus, and guided by prolific supervisors.
- The candidate will be able to contribute to the continued development of the BEAST code base, a software package used by many researcher in the fields of phylogenetics and molecular evolution, and collaborate with top researchers in the field.