A pharmacokinetic model based on the SSA-1DCNN-Attention method.

Area of Science:

• Pharmacokinetics
• Machine Learning
• Computational Chemistry

Background:

• Machine learning for pharmacokinetic (PK) indicator prediction often suffers from poor accuracy due to limited, unrepresentative training data.
• Developing robust predictive models is crucial for understanding drug behavior and optimizing therapeutic strategies.

Purpose of the Study:

• To address limitations in PK indicator prediction caused by small sample sizes and poor data representativeness.
• To propose and validate a novel 1DCNN-Attention model optimized by the sparrow search algorithm (SSA) for enhanced drug concentration prediction.

Main Methods:

• Data augmentation using the SMOTE (Synthetic Minority Over-sampling Technique) method to increase training set diversity and representativeness.
• Development of a 1DCNN-Attention model incorporating an attention mechanism to weigh variable importance for predicting drug concentration.
• Optimization of model parameters using the SSA algorithm to boost prediction accuracy.

Main Results:

• The proposed 1DCNN-Attention-SSA model demonstrated superior performance in predicting phenobarbital (PHB) concentrations in a specific epilepsy treatment model.
• The SMOTE method effectively expanded the small sample data, improving model generalizability.
• The attention mechanism successfully identified key pharmacokinetic indicators influencing drug concentration.

Conclusions:

• The SSA-optimized 1DCNN-Attention model offers a significant improvement for predicting pharmacokinetic indicators, especially with limited data.
• This approach enhances the reliability and accuracy of machine learning models in pharmaceutical research.
• The model's effectiveness was validated using a real-world case study involving PHB and epilepsy treatment.