Feature extraction approaches for biological sequences: a comparative study of mathematical features

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Genomic sequencing projects generate vast amounts of biological data.
• Machine learning performance in genomic sequence analysis is heavily dependent on feature extraction.
• Extracting significant discriminatory information from biological sequences is a key challenge.

Purpose of the Study:

• To propose and evaluate a new feature extraction pipeline using mathematical features.
• To analyze the effectiveness of Fourier, entropy, and complex network-based features for biological sequence classification.
• To assess the pipeline's performance and robustness across different RNA sequence classification tasks and imbalanced datasets.

Main Methods:

• Utilized mathematical features including Fourier transforms, entropy measures, and complex network analysis for numerical mapping of biological sequences.
• Developed a new feature extraction pipeline.
• Applied the pipeline to classify long non-coding RNA (lncRNA) and messenger RNA (mRNA) sequences.
• Validated the approach on predicting the class of lncRNA, such as circular RNAs.
• Tested the robustness of the method using imbalanced datasets.

Main Results:

• Conducted an in-depth study of various mathematical features for biological sequence representation.
• Introduced a novel feature extraction pipeline.
• Demonstrated high performance and robustness of the proposed pipeline in distinct RNA sequence classification tasks.
• Confirmed effectiveness even with imbalanced data scenarios.

Conclusions:

• The proposed feature extraction pipeline, leveraging mathematical features, offers a powerful approach for biological sequence classification.
• The method shows significant promise for handling large-scale genomic data and diverse classification challenges.
• The pipeline's robustness makes it suitable for real-world applications with potentially imbalanced biological datasets.