Feature Set Analysis of RNA-RNA Interaction Dataset

Understanding RNA-RNA interactions (RRIs) is fundamental to unraveling complex biological processes, including gene regulation, RNA splicing, translation, localization, and degradation. In this work, we present a comprehensive machine learning-based feature set analysis of RNA-RNA interaction datasets to identify features that contribute most significantly to accurate interaction prediction. By leveraging diverse machine learning (ML) methods, including feature selection algorithms and explainable AI tools, we systematically analyze multiple feature categories, such as sequence composition, secondary structure, thermodynamic properties, and base-pairing probabilities. Comparative experiments using classifiers such as Random Forests (RF), Support Vector Machines (SVM), Multi-Layer Perceptrons (MLP), and Convolutional Neural Networks (CNNs) demonstrate that optimized feature sets improve predictive performance while reducing computational overhead. Feature importance analysis highlights biologically relevant signals, including thermodynamic stability, conserved motifs, and structural flexibility, which are critical for RNA interaction prediction. Furthermore, the integration of SHAP (SHapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) enhances model interpretability, allowing alignment of computational predictions with biological reasoning. This work provides a thorough methodological and biological framework for RRI prediction, supporting downstream applications such as RNA-targeted therapeutics, functional annotation of non-coding RNAs, and transcriptome-wide interaction mapping. The presented pipeline serves as a robust foundation for developing interpretable, efficient, and biologically meaningful RNA-RNA interaction prediction tools.