piRNAPred

piRNAPred predicts PIWI-interacting RNAs (piRNAs), 21–35 nucleotide small non-coding RNAs, from sequence data to enable identification of molecules involved in gene expression regulation, transposon silencing, and antiviral responses.

Key Features:

• Hybrid Feature Integration: Uses k-mer nucleotide composition, predicted secondary structure, thermodynamic properties, and physicochemical characteristics as input features.
• Dataset Utilization: Trains and evaluates models on a non-redundant dataset of 1684 experimentally verified piRNAs and 1665 non-piRNA sequences sourced from piRBase and NONCODE.
• Machine Learning Approach: Applies various machine learning techniques for model development, with support vector machines (SVM) reported as the best-performing classifier.
• Performance Metrics: Reports ten-fold cross-validation results with 98.60% overall accuracy, Matthews correlation coefficient (MCC) of 0.97, and receiver operating characteristic (ROC) score of 0.99.
• Dimensionality Reduction: Implements attribute selection classifiers to reduce feature dimensionality and improve computational efficiency.

Scientific Applications:

• piRNA discovery: Predicts novel piRNAs to expand the annotated piRNA repertoire.
• Gene regulation studies: Enables investigation of piRNA roles in gene expression regulation.
• Transposon silencing research: Supports analysis of piRNA-mediated transposon control mechanisms.
• Antiviral response research: Facilitates study of piRNA involvement in viral infection inhibition.

Methodology:

Extracts k-mer composition, predicted secondary structure, thermodynamic and physicochemical features; applies attribute selection classifiers for dimensionality reduction; trains models using various machine learning techniques including support vector machines (SVM); and evaluates performance by ten-fold cross-validation on a non-redundant dataset of 1684 piRNAs and 1665 non-piRNAs from piRBase and NONCODE.