mSRFR: a machine learning model using microalgal signature features for ncRNA classification.

Area of Science:

• Bioinformatics
• Molecular Biology
• Genomics

Background:

• Accurate classification of noncoding RNAs (ncRNAs) is crucial for understanding gene regulation in microalgae.
• Existing tools may face challenges with data imbalance and feature selection specific to microalgal ncRNAs.

Purpose of the Study:

• To develop and validate a novel classification tool, mSRFR (microalgae SMOTE Random Forest Relief model), for identifying ncRNAs in diverse microalgae species.
• To address data imbalance and identify significant sequence and structure-based features for improved ncRNA prediction.

Main Methods:

• Applied SMOTE (Synthetic Minority Over-sampling Technique) to handle imbalanced datasets from the EBI RNA-central database.
• Utilized the Relief feature selection method to identify the top 20 significant features from 106 sequence and structure-based features.
• Employed ten-fold cross-validation to select the optimal classifier, with Random Forest demonstrating the highest performance (ROC area of 0.992).

Main Results:

• The mSRFR model achieved high accuracy (approximately 97%) and a low false-positive rate (approximately 2%) on a microalgal test dataset.
• Compared to other tools (RNAcon, CPC, CPC2, CNCI, CPPred), mSRFR showed superior predictive performance.
• The Relief analysis identified the %GA dinucleotide as a distinctive signature feature of microalgal ncRNAs compared to other organisms.

Conclusions:

• The mSRFR model provides an accurate and efficient method for classifying microalgal ncRNAs.
• The identification of the %GA dinucleotide as a signature feature offers new insights into microalgal ncRNA biology.
• This tool has significant implications for microalgal genomics and functional RNA research.