Feature extraction

Feature extraction extracts discriminatory numerical and network-based features from biological sequences to support machine-learning classification and analysis of RNAs such as long non-coding RNAs (lncRNAs), mRNAs, and circular RNAs.

Key Features:

• Mathematical feature extraction: Uses numerical mapping with Fourier transforms, entropy calculations, and complex network analyses to derive discriminatory features from biological sequences.
• lncRNA and mRNA focus: Applies feature design and assessment specifically to long non-coding RNAs (lncRNAs) and mRNAs.
• Validation across classification problems: Validated the extracted features on multiple classification tasks, including prediction of circular RNAs.
• Robustness to imbalanced data: Assesses performance and robustness under class-imbalanced scenarios.
• Feature extraction pipeline: Introduces a pipeline integrating the mathematical features to improve sequence classification performance.
• In-depth evaluation of mathematical features: Conducts detailed analysis of several mathematical features tailored for biological sequence analysis.

Scientific Applications:

• RNA sequence classification: Enables classification of RNA types including lncRNAs, mRNAs, and circular RNAs using extracted features.
• Genomic data analysis: Provides mathematical features that can be applied to genomic datasets to derive informative sequence descriptors for downstream analyses.

Methodology:

Computational methods explicitly include numerical mapping, Fourier transforms, entropy calculations, and complex network analyses; development and evaluation were structured as three studies: assessment with lncRNA and mRNA, validation across classification problems (including circular RNA prediction), and analysis of robustness under imbalanced data conditions.