POSMM

POSMM classifies metagenomic short reads using alignment-free single-order Markov models and logistic regression to assign taxonomic labels from Phylum to Species with associated confidence scores.

Key Features:

• Alignment-free classification: Uses an alignment-free approach based on Markov models for sequence classification.
• Read length support: Handles short reads typically ranging from 30 to 500 nucleotides.
• Single-order Markov models: Employs single-order Markov models to compute sequence probabilities.
• Logistic regression scoring: Transforms raw Markov model probabilities into scores using logistic regression models.
• Python sklearn optimization: Generates and optimizes logistic regression models with the Python sklearn library.
• Taxonomic resolution: Assigns classifications across Phylum, Class, Order, Family, Genus, and Species ranks.
• Confidence reporting: Provides an associated confidence level for each classification.
• Database-free model generation: Builds models directly from genome fasta files during each run rather than relying on pre-built databases.
• Complementary integration: Can be used alongside ultrafast classifiers such as Kraken2 to improve overall metagenomic classification accuracy.
• High sensitivity for large datasets: Maintains high sensitivity when applied to large-scale whole genome or metagenome datasets.

Scientific Applications:

• Taxonomic profiling: Taxonomic classification of metagenomic reads from Phylum to Species for microbial community analysis.
• Whole-genome/metagenome analysis: Sensitive classification in large-scale whole genome and metagenome sequencing projects.
• Hybrid workflows: Paired use with ultrafast classifiers (e.g., Kraken2) to enhance accuracy in metagenomic sequence classification.
• Custom model generation: Generation of run-specific models from genome fasta files to tailor classification to available reference genomes.

Methodology:

POSMM computes sequence probabilities using single-order Markov models, converts those probabilities into scores via logistic regression models generated and optimized with the Python sklearn library, thresholds the resulting scores for taxonomic assignment, and constructs models from input genome fasta files at run time.