Couplet Supertree based species tree Estimation

Abstract

Inference of a species tree from multi-locus gene trees having topological incongruence due to incomplete lineage sorting (ILS), is currently performed by either consensus (supertree), parsimony analysis (minimizing deep coalescence), or statistical methods. However, statistical approaches involve huge computational complexity. Accuracy of approximation heuristics used in either consensus or parsimony analysis, also varies considerably. We propose COSPEDSpec, a novel two stage species tree estimation method, combining both consensus and parsimony approaches. First stage uses our earlier proposed couplet supertree technique COSPEDTree [2] [3], whereas the second stage proposes a greedy heuristic to refine a non-binary (unresolved) supertree into a binary species tree. During each iteration, it reduces the number of extra lineages between the current species tree and the input gene trees, thus modeling ILS as the cause of gene tree / species tree incongruence. COSPEDSpec incurs time and space complexity lower or equal to the reference methods. For large scale datasets having hundreds of taxa and thousands of gene trees, COSPEDSpec produces species trees with lower branch dissimilarities and much less computation.