Scholarlink

Purpose <p>Accurate prediction of human clearance (CL) is essential in early drug development. Single Species Scaling (SSS) using rat pharmacokinetic (PK) data, particularly with unbound plasma fraction (f<sub>u,plasma</sub>), is widely used. However, its accuracy declines for compounds with extremely low f<sub>u,plasma</sub>, and no systematic method has addressed this limitation.</p> Methods <p>We developed a new approach, called Fraction-based Linear EXtrapolation SSS (FLEX-SSS fu Rat), which switches between SSS fu Rat and SSS Rat formulas based on an optimized fu threshold. The threshold and scaling coefficients were derived using a training set of 200 compounds. Additionally, a random forest (RF) machine learning model was built using molecular descriptors. Both models were validated using an external dataset of 62 compounds.</p> Results <p>All five predictive models showed comparable performance; among them, the consensus model combining FLEX-SSS fu Rat and RF yielded the best result: 40.3% within 2-fold error, only 16.1% above 5-fold, and GMFE of 2.7.</p> Conclusion <p>This study is the first to systematically validate SSS fu Rat using an independent dataset. The integration of threshold-based allometry and machine learning enabled more accurate human CL prediction, supporting informed decisions in first-in-human dose selection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fraction-based Linear Extrapolation (FLEX) Method for Predicting Human Pharmacokinetic Clearance: Advanced Allometric Scaling Method and Machine Learning Approach

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Yuki Umemori,
  • Koichi Handa,
  • Saki Yoshimura,
  • Michiharu Kageyama

摘要

Purpose

Accurate prediction of human clearance (CL) is essential in early drug development. Single Species Scaling (SSS) using rat pharmacokinetic (PK) data, particularly with unbound plasma fraction (fu,plasma), is widely used. However, its accuracy declines for compounds with extremely low fu,plasma, and no systematic method has addressed this limitation.

Methods

We developed a new approach, called Fraction-based Linear EXtrapolation SSS (FLEX-SSS fu Rat), which switches between SSS fu Rat and SSS Rat formulas based on an optimized fu threshold. The threshold and scaling coefficients were derived using a training set of 200 compounds. Additionally, a random forest (RF) machine learning model was built using molecular descriptors. Both models were validated using an external dataset of 62 compounds.

Results

All five predictive models showed comparable performance; among them, the consensus model combining FLEX-SSS fu Rat and RF yielded the best result: 40.3% within 2-fold error, only 16.1% above 5-fold, and GMFE of 2.7.

Conclusion

This study is the first to systematically validate SSS fu Rat using an independent dataset. The integration of threshold-based allometry and machine learning enabled more accurate human CL prediction, supporting informed decisions in first-in-human dose selection.