Literature mining on pharmacokinetics numerical data: a feasibility study.

Area of Science:

• Pharmacokinetics
• Computational Biology
• Data Mining

Background:

• Accurate pharmacokinetic (PK) parameter data is crucial for drug development and clinical decision-making.
• Existing literature mining methods often struggle with extracting complex numerical PK data efficiently and accurately.

Purpose of the Study:

• To develop and validate a sequential literature mining strategy for extracting numerical drug PK parameters.
• To compare the performance of the novel approach against conventional data mining techniques.

Main Methods:

• Building an entity template library to identify relevant pharmacokinetic articles.
• Applying tagging and extraction rules to retrieve PK data from abstracts.
• Utilizing a linear mixed meta-analysis model and an E-M algorithm for PK parameter estimation.
• Implementing a cross-validation procedure to assess false-positive rates.

Main Results:

• The sequential mining approach achieved 88% precision, 92% recall, and a 90% F-score for midazolam (MDZ) PK data.
• This performance significantly surpasses the 68.1% F-score of a conventional support vector machine approach.
• Comparable high performance was observed across investigations of seven additional drugs.

Conclusions:

• The developed sequential literature mining strategy is a feasible and highly effective method for extracting drug PK data.
• This approach offers a substantial improvement over traditional data mining techniques for PK parameter analysis.
• The method demonstrates robust performance across multiple drug PK datasets.