DCiPatho: deep cross-fusion networks for genome scale identification of pathogens

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Accurate pathogen detection is crucial for global disease control.
• Current deep learning methods struggle with long genomic sequences.
• Advancing pathogen identification requires novel computational approaches.

Purpose of the Study:

• To develop an advanced deep learning tool for accurate pathogen detection from genomic data.
• To overcome limitations of existing algorithms in processing long sequences.
• To enhance the identification of diverse pathogenic bacteria across different hosts.

Main Methods:

• Developed DCiPatho, a deep cross-fusion network integrating cross, residual, and deep neural networks.
• Utilized integrated frequency features of 3-to-7 k-mers for analysis.
• Evaluated performance on genomics and metagenomics datasets, including unlearned pathogen species.

Main Results:

• DCiPatho demonstrated accurate identification of distinct pathogenic bacteria affecting humans, animals, and plants.
• The tool effectively processed long genomic sequences, outperforming existing state-of-the-art algorithms.
• Successful evaluation on both known and novel pathogen species confirmed robustness.

Conclusions:

• DCiPatho is an effective tool for genomic-scale pathogen identification.
• Integrating k-mer frequencies into deep cross-fusion networks enhances detection accuracy.
• The developed method offers a significant advancement in computational pathogen diagnostics.