Ancestral population genomics using coalescence hidden Markov models and heuristic optimisation algorithms. (August 2015)

• Record Type: Journal Article
• Authors: Cheng, Jade Yu; Mailund, Thomas
• Is Part Of: Computational biology and chemistry. Volume 57(2015)
• Page Start: 80
• Page End: 92
• Publication Date: 2015-08
• Subjects: Sequential Markov coalescence -- Coalescent hidden Markov models -- Demographic inference -- Numerical optimisation -- Genetic algorithm -- Particle swarm optimisation; Chemistry -- Data processing -- Periodicals; Biology -- Data processing -- Periodicals; Biochemistry -- Data processing; Biology -- Data processing; Molecular biology -- Data processing; Periodicals; Electronic journals; 542.85
• DOI: 10.1016/j.compbiolchem.2015.02.001
• Languages: English
• ISSNs: 1476-9271

Abstract

Abstract: With full genome data from several closely related species now readily available, we have the ultimate data for demographic inference. Exploiting these full genomes, however, requires models that can explicitly model recombination along alignments of full chromosomal length. Over the last decade a class of models, based on the sequential Markov coalescence model combined with hidden Markov models, has been developed and used to make inference in simple demographic scenarios. To move forward to more complex demographic modelling we need better and more automated ways of specifying these models and efficient optimisation algorithms for inferring the parameters in complex and often high-dimensional models. In this paper we present a framework for building such coalescence hidden Markov models for pairwise alignments and present results for using heuristic optimisation algorithms for parameter estimation. We show that we can build more complex demographic models than our previous frameworks and that we obtain more accurate parameter estimates using heuristic optimisation algorithms than when using our previous gradient based approaches. Our new framework provides a flexible way of constructing coalescence hidden Markov models almost automatically. While estimating parameters in more complex models is still challenging we show that using heuristic optimisation algorithms we still get a fairly good accuracy.