Adaptive one-class Gaussian processes allow accurate prioritization of oncology drug targets

Area of Science:

• Computational biology
• Drug discovery and development
• Machine learning in pharmacology

Background:

• The escalating costs of research and development (R&D) in the pharmaceutical industry necessitate more efficient drug discovery processes.
• Accurate selection and prioritization of drug targets are critical for successful and cost-effective drug development.
• Traditional methods face challenges due to the lack of negative samples in target identification.

Purpose of the Study:

• To present a novel Gaussian Process model for prioritizing drug targets.
• To address the challenge of learning with only positive and unlabeled examples in drug target selection.
• To develop a robust method for hyperparameter selection in One-Class Gaussian Processes.

Main Methods:

• Utilized a Gaussian Process model framed as a learning problem with positive and unlabeled data.
• Developed a novel approach for hyperparameter selection of the kernel in One-Class Gaussian Processes.
• Applied the model to predict the druggability of oncology drug targets using a benchmark dataset.

Main Results:

• Achieved an Area Under the Curve (AUC) of 0.90 in predicting oncology drug targets from a small training set.
• The model successfully recovered a majority of known drug targets.
• Demonstrated the capability to identify novel protein candidates for drug targeting.

Conclusions:

• The proposed Gaussian Process model offers an effective solution for drug target prioritization, particularly in data-scarce scenarios.
• This approach can significantly improve the efficiency and reduce the cost of drug discovery.
• The model holds potential for identifying new therapeutic targets and advancing oncology drug development.