NGS read classification using AI

Area of Science:

• Genomics
• Bioinformatics
• Infectious Disease Diagnostics

Background:

• Clinical metagenomics offers broad pathogen detection but faces challenges in identifying novel or database-absent microbes.
• Current diagnostic methods rely on comparing sequencing data to reference databases, which can lead to false negatives for uncatalogued pathogens.

Purpose of the Study:

• To develop a novel computational method for detecting novel pathogens in metagenomic data.
• To address the limitation of reference database dependency in clinical metagenomics.

Main Methods:

• A neural network was trained on protein-coding sequences labeled by taxonomic domain.
• The trained neural network was used to classify unclassified sequences from metagenomic datasets.
• This approach facilitates the detection of potential novel pathogens by analyzing protein-level data.

Main Results:

• The study presents a new method for classifying protein sequences within metagenomic data.
• This method aids in identifying pathogens that may not be present in existing reference databases.
• The approach enhances the diagnostic capabilities of clinical metagenomics for novel infectious agents.

Conclusions:

• The developed neural network approach effectively classifies protein sequences, aiding in the detection of novel pathogens.
• This method overcomes the limitations of traditional database-comparison techniques in metagenomic analysis.
• This advancement holds promise for improving infectious disease diagnostics and discovering new microbial threats.