ProFET: Feature engineering captures high-level protein functions.

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• The rapid growth of sequenced genomes and proteins outpaces manual functional annotation.
• Homologous inference methods struggle with diverse protein functions and broad classifications.
• Accurate identification of high-level protein functionality remains a significant challenge.

Purpose of the Study:

• To develop a universal feature engineering approach for classifying high-level protein functions.
• To create a bioinformatics toolkit that combines feature engineering with machine learning.
• To enable functional classification without reliance on external databases or sequence alignment.

Main Methods:

• Development of the Protein Feature Engineering Toolkit (ProFET).
• Extraction of hundreds of biophysical and sequence-derived features.
• Implementation of ProFET in Python for multi-genome scale analysis.
• Application on 17 benchmark datasets for binary and multi-class classification tasks.

Main Results:

• ProFET achieved state-of-the-art performance across various protein classification tasks.
• Extracted features demonstrated excellent biological interpretability.
• Successful application to diverse high-level functions including subcellular localization, structural classes, and specific properties like thermophilic and nucleic acid binding.

Conclusions:

• ProFET offers a powerful and universal method for protein feature engineering and functional classification.
• The toolkit facilitates the discovery of novel target proteins and understanding of functional properties.
• ProFET integrates seamlessly into data analysis pipelines for large-scale genomic studies.