NeRNA: A negative data generation framework for machine learning applications of noncoding RNAs.

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Molecular Biology
• Machine Learning in Biology

Background:

• Supervised machine learning methods for noncoding RNA (ncRNA) analysis require high-quality training datasets.
• Existing datasets often lack validated negative examples for specific ncRNA classes, hindering model development.
• The absence of standardized methods for generating negative ncRNA data presents a significant challenge.

Purpose of the Study:

• To introduce NeRNA, a novel method for generating high-quality negative RNA sequence data.
• To address the limitations of negative data availability in ncRNA machine learning.
• To evaluate the performance of NeRNA-generated datasets across various ncRNA types and machine learning models.

Main Methods:

• NeRNA utilizes known ncRNA sequences and their structures for octal representation to create artificial negative sequences.
• The method mimics frameshift mutations without introducing insertions or deletions.
• NeRNA was tested on microRNA (miRNA), transfer RNA (tRNA), long noncoding RNA (lncRNA), and circular RNA (circRNA) datasets using various classifiers and deep learning models.

Main Results:

• Machine learning models trained with NeRNA-generated datasets demonstrated substantially high prediction performance across multiple ncRNA types.
• Cross-validation results (1000-fold) confirmed the effectiveness of NeRNA for miRNA prediction, including species-specific analyses.
• The performance was consistent across diverse classifiers (Decision Tree, Naïve Bayes, Random Forest) and deep learning architectures (MLP, CNN, Simple feedforward NN).

Conclusions:

• NeRNA effectively overcomes the challenge of generating high-quality negative RNA sequence data for machine learning.
• The method significantly enhances the predictive accuracy of ncRNA classification models.
• NeRNA is provided as an accessible KNIME workflow, facilitating its adoption in RNA sequence data analysis.