PC-mer: An Ultra-fast memory-efficient tool for metagenomics profiling and classification

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• K-mer-based methods are crucial for metagenomics data analysis but face performance and memory limitations.
• Existing feature extraction techniques present bottlenecks in processing large-scale biological sequence data.

Purpose of the Study:

• To develop an innovative feature extraction and sequence profiling method for DNA/RNA sequences.
• To overcome the limitations of k-mer methods in metagenomics classification and analysis.

Main Methods:

• Developed PC-mer, a novel method utilizing physicochemical properties of nucleotides for feature extraction.
• Compared PC-mer with traditional k-mer profiling methods on various machine learning and computational approaches.

Main Results:

• PC-mer reduces memory usage by a factor of 2k compared to k-mer methods.
• Achieved over 1000x speedup in the training phase for metagenomics classification.
• Demonstrated 100% accuracy in classifying samples at class, order, and family levels.
• Improved genus-level classification accuracy by >14% (shotgun) and >5% (amplicon) datasets.

Conclusions:

• PC-mer offers a significant advancement over k-mer methods for metagenomics data analysis.
• The method provides substantial improvements in memory efficiency, speed, and classification accuracy.
• Introduced two PC-mer-based tools for classifying and comparing metagenomics data, offering viable alternatives to k-mer tools.