Machine Learning-Based Prediction of Reduction Potentials for Pt^IV Complexes.

Area of Science:

• Computational Chemistry
• Medicinal Chemistry
• Machine Learning in Drug Discovery

Background:

• Platinum(IV) complexes offer advantages over platinum(II) drugs as inert prodrugs.
• Their activation relies on reduction to active platinum(II) species, with reduction potential being a key factor.
• Understanding and predicting reduction potentials is vital for rational drug design.

Purpose of the Study:

• To develop a machine learning (ML) model for predicting the reduction potentials of platinum(IV) complexes.
• To identify key molecular descriptors influencing these electrochemical properties.
• To facilitate the rational design of novel platinum(IV) prodrugs with tailored pharmaceutical applications.

Main Methods:

• Utilized a dataset of experimentally determined reduction potentials for platinum(IV) complexes.
• Employed various machine learning algorithms and feature engineering techniques.
• Incorporated ab initio calculations and diverse molecular descriptors (constitutional, topological, electronic).

Main Results:

• Achieved high predictive accuracy for reduction potentials (R² = 0.92, RMSE = 0.13 V).
• Identified 2D Atom Pairs descriptors and lowest unoccupied molecular orbital energy as significant features.
• Demonstrated that a select set of 20 descriptors can effectively differentiate complexes by reduction potential.

Conclusions:

• The ML approach provides a rapid and effective tool for predicting platinum(IV) complex reduction potentials.
• This method significantly aids in the rational design and screening of novel platinum(IV) prodrug candidates.
• The findings offer valuable insights into structure-property relationships for electrochemical applications in medicine.