NRPreTo: A Machine Learning-Based Nuclear Receptor and Subfamily Prediction Tool

Area of Science:

• Biochemistry
• Bioinformatics
• Genomics

Background:

• The nuclear receptor (NR) superfamily comprises ligand-activated proteins crucial for cellular activities.
• NRs are classified into seven subfamilies, and identifying them aids in understanding their functions and disease associations.
• Existing NR prediction tools have limitations due to reliance on limited sequence-based features, potentially causing overfitting.

Purpose of the Study:

• To develop a robust, two-level prediction tool for identifying nuclear receptors (NRs) and their subfamilies.
• To improve NR prediction accuracy by incorporating a wider range of protein features beyond sequence-based data.
• To provide insights into functional relationships and disease pathways involving NRs.

Main Methods:

• Developed Nuclear Receptor Prediction Tool (NRPreTo), a two-level prediction system.
• Utilized sequence-based features along with six additional feature groups (physiochemical, structural, evolutionary).
• Employed Random Forest classifiers on benchmark, RefSeq, and HPRD human protein datasets.

Main Results:

• The incorporation of additional feature groups significantly enhanced prediction performance.
• NRPreTo demonstrated high accuracy on external datasets.
• The tool successfully predicted 59 novel nuclear receptors within the human proteome.

Conclusions:

• NRPreTo offers a more accurate and comprehensive approach to NR identification and subfamily classification.
• The tool's enhanced feature set overcomes limitations of existing prediction methods.
• NRPreTo facilitates deeper understanding of NR roles in biological processes and diseases.