CMStatistics 2018: Start Registration
View Submission - CMStatistics
B0687
Title: Approximate message passing algorithms for de novo reconstruction in metagenomics Authors:  Sergio Bacallado - Cambridge University (United States) [presenting]
Abstract: Microbiome studies sequence the DNA in samples containing a mixture of bacterial genomes. DNA sequences must be assigned to different taxa, and when not all taxa have been cultured and characterised previously this problem is known as de novo reconstruction. Mathematically, de novo reconstruction is a deconvolution problem which reduces to a matrix factorisation with highly structured factors. As it is relatively easy to formulate a prior and a probability model, Bayesian approaches to this problem have been proposed, but their computational cost can be high. We present an alternative algorithm based on approximate message passing which is evaluated by simulation.