DL-TODA: A Deep Learning Tool for Omics Data Analysis.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Metagenomics enables genome-wide microbiome profiling by analyzing DNA sequences (reads).
• Efficient computational tools are crucial for classifying vast amounts of metagenomic data without reference databases.
• Existing tools like Kraken2 and Centrifuge are state-of-the-art but can be improved for accuracy and bias reduction.

Purpose of the Study:

• To introduce DL-TODA, a novel deep learning program for metagenomic read classification.
• To evaluate DL-TODA's performance against established tools like Kraken2 and Centrifuge.
• To demonstrate DL-TODA's utility in diverse microbiome environments.

Main Methods:

• Utilized a deep learning model, specifically a convolutional neural network, trained on over 3000 bacterial species.
• Applied a computer vision architecture for modeling species-specific features.
• Tested DL-TODA using synthetic data from 2454 genomes (639 species) and real-world metagenomes (human oral, cropland soil).

Main Results:

• DL-TODA achieved high confidence classification for nearly 75% of synthetic reads.
• Classification accuracy exceeded 0.98 at taxonomic ranks above genus, comparable to Kraken2 and Centrifuge.
• Achieved 0.97 species-level accuracy, outperforming Kraken2 (0.93) and Centrifuge (0.85).
• DL-TODA demonstrated distinct relative abundance rankings and reduced bias in real-world microbiome samples.

Conclusions:

• DL-TODA is an effective deep learning tool for accurate metagenomic read classification.
• The program offers superior species-level accuracy and reduced taxonomic bias compared to current leading tools.
• DL-TODA is suitable for analyzing microbiomes across diverse environments.