Taxonomic classification with maximal exact matches in KATKA kernels and minimizer digests

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Taxonomic classification of DNA reads is crucial for understanding microbial communities.
• Maximal Exact Matches (MEMs) offer potential advantages over k-mers for accurate classification.
• Current MEM-based methods are limited by the computational cost associated with large genome datasets.

Purpose of the Study:

• To investigate the efficacy of lossy compression techniques for indexing genomes in a phylogenetic tree.
• To evaluate the performance of MEM-based classification using compressed genome representations.
• To assess the trade-off between compression ratio and classification accuracy.

Main Methods:

• Constructed an augmented FM-index over concatenated genomes.
• Utilized Maximal Exact Matches (MEMs) for identifying sequence origins.
• Applied three lossy compression methods: KATKA kernel, minimizer digest, and a combination of both.
• Simulated reads and evaluated true-positive rates across various parameter settings.

Main Results:

• Achieved significant data compression using KATKA kernels and minimizer digests.
• Demonstrated that lossy compression slightly decreases the true-positive rate for MEM-based classification.
• Identified parameter settings that balance compression and accuracy.

Conclusions:

• Lossy compression is a viable strategy for enabling efficient MEM-based taxonomic classification of large genomic datasets.
• The choice of compression technique and parameters impacts the trade-off between data size and classification performance.
• This approach holds promise for scalable and accurate genomic read classification.