DEPP: Deep Learning Enables Extending Species Trees using Single Genes.

Area of Science:

• Computational Biology
• Evolutionary Biology
• Bioinformatics

Background:

• Phylogenetic placement is crucial for microbiome and environmental sample analysis.
• Current methods rely on specific evolutionary models, limiting their application when models are unknown.
• Extending genome-wide species trees with single-gene data presents a significant challenge.

Purpose of the Study:

• To introduce Deep-learning Enabled Phylogenetic Placement (DEPP), an algorithm for extending species trees using single genes without prespecified evolutionary models.
• To demonstrate DEPP's accuracy and applicability in various phylogenetic analyses.
• To enable integrated microbiome analyses using diverse genetic data.

Main Methods:

• Developed Deep-learning Enabled Phylogenetic Placement (DEPP), a novel algorithm utilizing neural networks.
• Trained and validated DEPP on simulated and real biological datasets.
• Applied DEPP to update microbial tree-of-life and combine 16S and metagenomic data.

Main Results:

• DEPP achieves accuracy comparable to model-based methods without prior model knowledge.
• The algorithm accurately updates the multilocus microbial tree-of-life using single genes.
• DEPP successfully integrates 16S and metagenomic data for unified phylogenetic analysis.

Conclusions:

• DEPP offers a powerful, model-agnostic approach for phylogenetic placement and species tree extension.
• This method enhances the analysis of complex environmental samples, particularly microbiomes.
• DEPP facilitates more comprehensive community structure analyses by leveraging multiple data types.