MathFeature: feature extraction package for DNA, RNA and protein sequences based on mathematical descriptors

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Machine learning on biological sequences requires effective numerical representation.
• Existing feature extraction methods are often not readily available in standard packages.
• Mathematical descriptors are crucial for capturing sequence information.

Purpose of the Study:

• Introduce MathFeature, a novel package for extracting numerical features from biological sequences.
• Implement 20 mathematical descriptors based on diverse computational approaches.
• Enhance machine learning applications in genomics and proteomics.

Main Methods:

• Developed the MathFeature package implementing 20 mathematical descriptors.
• Utilized approaches including numeric mappings, genomic signal processing, chaos game theory, entropy, and complex networks.
• Validated descriptor robustness and relevance through nine case studies and eight benchmark datasets.

Main Results:

• MathFeature descriptors achieved high performance (0.6350-0.9897 accuracy) in case studies.
• Features extracted by MathFeature outperformed existing methods in benchmark datasets.
• Hybridization of MathFeature descriptors with known methods further improved performance.

Conclusions:

• MathFeature provides unique, robust mathematical descriptors for biological sequence analysis.
• The package enhances machine learning model performance for DNA, RNA, and protein sequences.
• MathFeature democratizes advanced feature extraction for researchers, including non-experts.