Predicting Coding Potential of RNA Sequences by Solving Local Data Imbalance.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Non-coding RNAs (ncRNAs) are vital in biological processes and disease.
• Accurate prediction of RNA coding potential is essential for functional analysis.
• Existing machine learning methods struggle with RNA sequences containing short Open Reading Frames (sORFs).

Purpose of the Study:

• To develop a novel method for predicting RNA coding potential, specifically addressing challenges posed by sORFs.
• To investigate and alleviate the problem of local data imbalance in datasets with sORFs.
• To improve the performance of coding potential prediction for RNA sequences with sORFs.

Main Methods:

• Analysis of ORF length distribution in RNA sequences.
• Proposal of the CPE-SLDI method incorporating data oversampling techniques.
• Augmentation of coding RNA samples with sORFs to address data imbalance.

Main Results:

• Identified a significant data imbalance issue, with fewer coding RNAs possessing sORFs compared to ncRNAs.
• Demonstrated that CPE-SLDI outperforms existing methods in coding potential prediction.
• Confirmed that data augmentation via oversampling enhances prediction performance, particularly for sORF-containing RNA sequences.

Conclusions:

• The CPE-SLDI method effectively mitigates local data imbalance for sORFs.
• Data augmentation strategies are crucial for improving coding potential prediction accuracy, especially for sequences with sORFs.
• The proposed method offers a more robust solution for distinguishing coding RNAs from ncRNAs, particularly in challenging cases.