MetaMLP: A Fast Word Embedding Based Classifier to Profile Target Gene Databases in Metagenomic Samples.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Metagenomic functional profiling is crucial for understanding microbial communities.
• Current methods like sequence alignment are computationally intensive.
• Alignment-free methods require substantial memory.

Purpose of the Study:

• To develop MetaMLP (Metagenomics Machine Learning Profiler), a novel machine learning approach for metagenomic functional profiling.
• To improve the speed and efficiency of metagenomic analysis.

Main Methods:

• MetaMLP represents sequences as numerical vectors (embeddings).
• It utilizes a one hidden layer neural network for functional category assignment.
• Employs a reduced alphabet for partial matching using full and partial k-mers.

Main Results:

• MetaMLP achieves high accuracy with 0.99 precision and 0.99 recall.
• It identifies more reads than DIAMOND, a fast alignment method.
• Processes 100 million reads in approximately 10 minutes on a laptop, 50x faster than DIAMOND.

Conclusions:

• MetaMLP offers a computationally efficient and accurate solution for metagenomic functional profiling.
• This method significantly accelerates the analysis of large-scale metagenomic datasets.
• Enables broader accessibility to advanced metagenomic analysis on standard hardware.