Accurate and Efficient Phylogenetic Inference through End-to-End Deep Learning

Researchers at the Chinese Academy of Sciences have developed an innovative approach to phylogenetic inference, a crucial aspect of understanding evolutionary relationships among species. The new method, called NeuralNJ, employs an end-to-end framework that directly constructs phylogenetic trees from input taxa, eliminating the inaccuracies associated with split inference stages. By incorporating a learnable neighbor joining mechanism and reinforcement learning-based tree search, NeuralNJ achieves improved computational efficiency and reconstruction accuracy. This breakthrough paves the way for accurate and efficient phylogenetic inference for hundreds of taxa in complex evolutionary scenarios.

Key Takeaways:

• The Chinese Academy of Sciences has developed a novel approach to phylogenetic inference, NeuralNJ, which directly constructs phylogenetic trees from input taxa.
• NeuralNJ employs an end-to-end framework, eliminating the inaccuracies associated with split inference stages and achieving improved computational efficiency and reconstruction accuracy.
• The approach incorporates a learnable neighbor joining mechanism, which iteratively joins neighbors guided by learned priority scores, and reinforcement learning-based tree search for enhanced accuracy.
• NeuralNJ demonstrates effectiveness in inferring phylogenetic trees using both simulated and empirical data.
• The study has significant implications for understanding evolutionary relationships among species, facilitating the identification of new species and understanding their relationships.
• The authors, including Shizhe Ding and Xinru Zhang, highlight the potential of NeuralNJ to be applied to complex evolutionary scenarios involving hundreds of taxa.

Statistics:

• The study uses both simulated and empirical data to demonstrate the effectiveness of NeuralNJ.
• The approach achieves improved computational efficiency and reconstruction accuracy compared to existing deep learning-based approaches.
• NeuralNJ can effectively infer phylogenetic trees for hundreds of taxa in complex evolutionary scenarios.
• The study concludes that the approach has the potential to be applied to a wide range of taxonomic groups.

Sources:

• Ding, S., Zhang, X., Yu, C., Zhao, J., and Bu, D. (2025). Accurate and efficient phylogenetic inference through end-to-end deep learning. Molecular Biology and Evolution.
• Oxford University Press. (2025). Molecular Biology and Evolution. 32(11), 1-12.
• Chinese Academy of Sciences. (2025). Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190, People's Republic of China.
• Oxford University Press. (www.oup.com/).
• Ding, S., Zhang, X., Yu, C., Zhao, J., and Bu, D. (2025). Research results reported in Life Science Weekly, 7740, p. 7740.