PredPS: Attention-based graph neural network for predicting stability of compounds in human plasma

Area of Science:

• Computational chemistry
• Drug discovery and development
• Bioinformatics

Background:

• Compound stability in human plasma is critical for effective drug delivery and in vivo efficacy.
• Rapid plasma degradation can lead to poor therapeutic outcomes.
• Existing tools for predicting human plasma stability are limited, with no open-source software available.

Purpose of the Study:

• To develop an open-source software program for predicting human plasma stability.
• To create an accurate and efficient in silico model for assessing compound degradation in plasma.
• To support early-stage drug discovery by providing a reliable stability prediction tool.

Main Methods:

• Development of an attention-based graph neural network model named PredPS.
• Training and validation using combined in-house and open-source datasets.
• Comparative analysis against existing machine learning and deep learning algorithms.

Main Results:

• PredPS demonstrated superior performance compared to other tested algorithms.
• Achieved high predictive accuracy with an Area Under the Receiver Operating Characteristic Curve (AUC) of 90.1%.
• Reported accuracy of 83.5%, sensitivity of 82.3%, and specificity of 84.6% via 5-fold cross-validation.

Conclusions:

• PredPS is an effective tool for predicting human plasma stability in early drug discovery.
• Adopting in silico plasma stability prediction models like PredPS can accelerate high-throughput screening.
• The availability of PredPS source code and a web server facilitates its adoption in research settings.