همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Predicting Plasma Protein Binding (Fraction Unbound) with Machine Learning Using Molecular Descriptors

نویسندگان :

Arash Maghsoudlou¹ Fatemeh Ghorbani-Bidkorpeh² M. Soltani³

1- School of Pharmacy, Shahid Beheshti University of Medical Sciences 2- School of Pharmacy, Shahid Beheshti University of Medical Sciences 3- University of Waterloo

کلمات کلیدی :

plasma protein binding،machine learning،fraction unbound،molecular descriptors

چکیده :

The unbound fraction in plasma (fu) is a key pharmacokinetic parameter that directly influences drug distribution, elimination, and in vivo efficacy. Experimental measurement of fu is costly and low-throughput, motivating the need for reliable computational prediction. In this study, we developed machine-learning models to predict fu using physicochemical descriptors and in vitro–derived molecular fingerprints extracted from the Krumpholz et al. dataset. Gradient Boosting achieved the best performance (R² ≈ 0.61, RMSE ≈ 0.19, MAE ≈ 0.15), outperforming Random Forest (R² ≈ 0.53). Feature-importance analysis revealed that lipophilicity- and polarity-related descriptors (e.g., NumHAcceptors, GST_max) were the dominant drivers of prediction. Overall, our findings demonstrate that ML models can capture the key molecular determinants of fu and offer a practical alternative to experimental assays.