IQ-NET: A Deep Learning Approach for Fast and Accurate Phylogenetic Inference

Chen Yang, Bui Quang Minh, Zixin Zhuang, Piyumal Anthony Demotte, Cuong Cao Dang, Le Sy Vinh, Nhan Ly-Trong

Keywords: Phylogenetic inference, Machine learning, Quartet analysis, Empirical data training

TL;DR: We present IQ-NET, a deep learning framework that reconstructs four-taxon phylogenies directly from MSAs.

Abstract: Phylogenetic inference reconstructs evolutionary relationships between species from molecular sequences. Although likelihood-based methods are rigorous, they are computationally expensive. Existing machine learning based phylogenetic approaches are faster but often trained on simulated data, which are far removed from real data, causing out-of-distribution predictions. Here, we introduce IQ-NET (Intelligent Quartet NETwork), a permutation-invariant deep learning framework trained directly on empirical alignments to estimate phylogenetic quartet trees. IQ-NET jointly infers the tree topology and branch lengths without model assumptions, achieving superior accuracy and a 24-fold speedup over the widely used IQ-TREE software.

Submission Number: 3