A machine learning approach for the identification of odorant binding proteins from sequence-derived properties

Area of Science:

• Biochemistry
• Bioinformatics
• Molecular Biology

Background:

• Odorant binding proteins (OBPs) are crucial for olfaction, transporting odorants to receptors.
• Predicting OBPs from sequence data is challenging due to low sequence similarity.
• Existing sequence-based prediction methods for OBPs are limited.

Purpose of the Study:

• To develop a novel algorithm for predicting OBPs using sequence-derived properties.
• To assess the efficacy of Regularized Least Squares Classifier (RLSC) for OBP prediction.
• To improve the identification of OBPs irrespective of sequence homology.

Main Methods:

• Developed a prediction algorithm utilizing Regularized Least Squares Classifier (RLSC).
• Incorporated multiple physicochemical properties of amino acids into the prediction model.
• Validated the algorithm on datasets from Pfam and GenDiS databases.

Main Results:

• Achieved an overall prediction accuracy of 97.7% (94.5% positive, 98.4% negative).
• Successfully predicted 92.8% of non-homologous OBPs and 97.1% of an independent dataset.
• Demonstrated high prediction accuracy even with low sequence similarity.

Conclusions:

• RLSC is a powerful tool for predicting OBPs from sequence-derived properties.
• The proposed method facilitates OBP identification, aiding olfactory research.
• The algorithm shows significant potential for discovering novel OBPs.