Nucleic acid sequence-based detection and characterization using genome-level pseudo-imaging and machine learning

BACKGROUND

Whole genome sequencing provides a complete snapshot of an organism’s genetic code, enabling evolutionary inference through quantification of nucleotide similarities. However, reference databases exhibit bias toward well-studied species due to long-tail distributions and sampling gaps. These imbalances risk species misidentification and hinder the characterization of novel taxa.

SUMMARY OF TECHNOLOGY

OSU Researchers have developed TEPI (Taxonomy-Aware Embedding and Pseudo-Imaging), revolutionizing genome classification through innovative embedding and imaging techniques for enhanced species recognition. Embedding space calculated by k-mer co-occurrences aids in the capture of hierarchical relationships between species and facilitates efficient and accurate species recognition through convolutional neural networks (CNNs) represented through pseudo-imaging.

POTENTIAL AREAS OF APPLICATION

• Genomic & Taxonomic Classification
• Computational Biology & Bioinformatics
• Antimicrobial Resistance Surveillance Networks
• Clinical Diagnostics
• Metagenomic Analysis

MAIN ADVANTAGES

• Process, Visualize, & Interpret Large-Scale Genomic Datasets
• Effective in Zero-Shot Scenarios
• Improved Computational Efficiency
• Scalability
• Genome Pseudo-Imaging
• Improved Genomic Similarity Predictions
• Addresses Model Bias

STAGE OF DEVELOPMENT

• Proof of Concept