QRAFT: Scaling coalescent-based species tree inference to one million taxa

QRAFT reconstructs a million-taxon species tree from 1,000 gene trees in three hours using 298 GB RAM and 64 CPU threads, without GPUs.

• Why it matters: Coalescent-based species tree inference has lagged behind phylogenomic datasets that now include hundreds of thousands of species, limiting analyses at tree-of-life scale.
• What they did: QRAFT maximizes quartet support under the multispecies coalescent and runs in near-linear time O(kn log^2 n) with balanced recursion, using asymptotically optimal O(kn) memory.
• The result: It achieves accuracy comparable to leading methods and, for a microbial tree of life with 199,330 genomes, reconstructs it in a few hours on modest CPU resources.

AI summary of the abstract · may contain errors · not medical advice · read the paper.