A novel riboswitch classification based on imbalanced sequences achieved by machine learning.

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Riboswitches are regulatory mRNA elements with aptamer and expression platform components.
• Classifying riboswitches is challenging due to imbalanced datasets, where one class has significantly fewer sequences than others.
• Imbalanced data can lead to biased classifiers that favor majority classes, compromising accuracy.

Purpose of the Study:

• To develop and evaluate a method for classifying riboswitches from imbalanced sequence data.
• To address the challenge of majority bias and overfitting in riboswitch classification algorithms.
• To identify sequence features with biological relevance for riboswitch classification.

Main Methods:

• Considered sixteen imbalanced riboswitch families for analysis.
• Utilized a novel pipeline for splitting sequences into training and testing sets.
• Extracted 5460 k-mers (k=1-6) and selected 156 features using CfsSubsetEval and BestFirst algorithms in WEKA 3.8.

Main Results:

• Statistically significant differences were observed between balanced and imbalanced sequence classifications (p < 0.05).
• Algorithms showed significant variations in sensitivity, specificity, accuracy, and macro F-score across balanced and imbalanced groups (p < 0.05).
• Identified k-mers with biological functions and motifs in riboswitch structures (loops, helices), validating their importance.

Conclusions:

• Solving majority bias and overfitting is crucial for accurate riboswitch classification.
• The developed models demonstrate generalized evaluation capabilities for classifying novel riboswitches.
• The study provides a robust approach to handle imbalanced data in regulatory mRNA sequence analysis.