Applying Neural Network to Reconstruction of Phylogenetic Tree

Reconstruction of phylogenetic tree from biological sequences is a fundamental step in molecular biology, but it is computationally exhausting. Our goal is to use neural network to learn the heuristic strategy of phylogenetic tree reconstruction algorithm. We propose an attention model to learn heuristic strategies for constructing circular ordering related to phylogenetic trees. We use alignment-free K-mer frequency vector representation to represent biological sequences and use unlabeled sequence data sets to train attention model through reinforcement learning. Comparing with traditional methods, our approach is alignment-free and can be easily extended to large-scale data with computational efficiency. With the rapid growth of public biological sequence data, our method provides a potential way to reconstruct phylogenetic tree.