Fast algorithms for inferring gene-species associations

Abstract

Assessment of microbial biodiversity is typically made by sequencing either PCR-amplified marker genes or all genomic DNA from environmental samples. Both approaches rely on the similarity of the sequenced material to known entries in sequence databases. However, amplicons of non-marker genes are often used, when the research question aims at assessing both functional capabilities of a microbial community and its biodiversity. In such cases, a phylogenetic tree is constructed with known and metagenomic sequences, and expert assessment defines the taxonomic groups the amplicons belong to. Here, instead of relying on sequences, often missing, of non-marker genes, we use tree reconciliation to obtain a distribution of mappings between genes and species. We describe efficient algorithms for the reconstruction of gene-species mappings and a Monte-Carlo method for the inference of distributions for the cases when the number of optimal reconstructions is large. We provide a comparative study of different cost functions showing that the duplication-loss cost induces mappings of the highest quality. Further, we demonstrate the correctness of our approach using several datasets.