A bi-layer model for identification of piwiRNA using deep neural learning

Area of Science:

• Biochemistry
• Genetics
• Bioinformatics

Background:

• PIWI-interacting RNAs (piRNAs) are non-coding RNAs crucial for gene regulation at transcriptional and post-transcriptional levels.
• They play vital roles in gametogenesis, transposon silencing, viral defense, and maintaining animal fertility.
• Accurate piRNA prediction is essential for understanding these cellular functions.

Purpose of the Study:

• To develop a highly accurate predictor for piRNA identification.
• To improve upon existing piRNA prediction methods.
• To explore novel feature descriptors and machine learning models for enhanced prediction.

Main Methods:

• Feature extraction using reverse complement k-mer, gapped k-mer composition, and k-mer composition.
• Feature selection employing cascade and relief strategies.
• Model training and validation using Random Forest (RF), Deep Neural Network (DNN), and Support Vector Machine (SVM) with 10-fold cross-validation.

Main Results:

• The Deep Neural Network (DNN) model, utilizing optimal features selected by the Cascade approach, achieved the highest prediction accuracy.
• The developed predictor, BLP-piwiRNA, demonstrated superior performance compared to existing piRNA prediction tools.
• The study identified effective feature sets for piRNA prediction.

Conclusions:

• BLP-piwiRNA represents a significant advancement in piRNA prediction accuracy.
• The findings offer valuable tools for the research community and the drug development industry.
• BLP-piwiRNA holds promise as potential biomarkers and therapeutic targets for cancer diagnostics and treatment.