isheshukov/bioinf-sem-project

Bayesian optimization for demographic history inference

Bayesian optimization for demographic history inference

Bioinformatics Institute, 2019

Project goals

The project goal was to replace a BFGS optimization algorithm used in the tool moments with the Gaussian process based global Bayesian optimization and to study the effects.

Methods

We used GPyOpt library that, given function and bounding conditions optimizes said function, then we took the optimization routine in moments and replaced their algorithm with the calls to GPyOpt and some other of our changes to make them work together.

To determine next query point for bayesian optimization we used Expected Improvement acquisition function, but it may be a subject for future rethinking.

Requirements

Python >= 2.5

(Tested on Ubuntu 19.04, Python 2.7.15)

How to run

Dependencies

It's best to use isolated environments like virtualenv or conda.

Install moments according to the instructions, then execute

pip install -r requirements.txt

to install dependencies.

Running

Run python2 2pop_6.py, python2 2pop_8.py or python2 3pop_13.py.

There are no parameters to set in CLI. You can change the program by editing the source code.

Clarification

Scripts (2pop_6.py, 2pop_8.py, 3pop_13.py) used employ standard Out of Africa hypothesis of human origin. They differ only in number of populations used (2 and 3 respectively) and number of parameters that a given model uses (6, 8 or 13).

Scripts try to produce the optimal parameters for each model that describe the demographics history for given population group.

Results

After the successful run of the one of the aforementioned scripts you'll see the results in out folder. They are in the form of csv table log.txt, GPyOpt reports and GPyOpt plots provided that your problem is 1- or 2-dimensional.

From the log file convergence plot may be constructed, for example