MP3

MP3 predicts pathogenic (virulent) proteins in genomic and metagenomic datasets to estimate pathogenic potential and detect partial pathogenic proteins from short metagenomic reads.

Key Features:

• Prediction target: Predicts pathogenic (virulent) proteins in both genomic and metagenomic datasets, including novel and unannotated sequences.
• Short-read detection: Identifies partial pathogenic proteins from short metagenomic reads (100-150 bp).
• Algorithms: Employs an integrated approach combining Support Vector Machine (SVM) and Hidden Markov Model (HMM) methodologies.
• Performance on complete proteins: Achieved Sensitivity 92%, Specificity 100%, Matthews Correlation Coefficient (MCC) 0.92, and accuracy 96% on a blind dataset of complete proteins.
• Performance on metagenomic fragments (Blind A): For 51–100 amino acid sequences, Sensitivity 82.39%, Specificity 97.86%, MCC 0.80, and accuracy 89.32%.
• Performance on metagenomic fragments (Blind B): For 30–50 amino acid sequences, Sensitivity 71.60%, Specificity 94.48%, MCC 0.67, and accuracy 81.86%.
• Speed and sensitivity: Uses the integrated SVM+HMM approach to provide fast, sensitive, and accurate predictions at scale.
• Validation: Validated on selected bacterial genomic and real metagenomic datasets.

Scientific Applications:

• Virulent protein identification: Identification of virulent proteins in newly sequenced genomes and unannotated genomic sequences.
• Pathogenic potential estimation: Estimation of pathogenic potential of genomes and metagenomes.
• Comparative metagenomics: Comparison of metagenomes from healthy and diseased individuals to assess the proportion of pathogenic species.
• Metagenomic read analysis: Detection of partial pathogenic proteins in analyses of short metagenomic reads (100-150 bp).

Methodology:

Combines Support Vector Machine (SVM) and Hidden Markov Model (HMM) methodologies and was evaluated on blind datasets of complete proteins and metagenomic fragments (Blind A: 51–100 aa; Blind B: 30–50 aa).