A Closed‐Loop Hybrid Discovery System of Type I Photosensitizers for Hypoxic Tumor Therapy

TL;DR

This paper introduces a system combining machine learning and calculations to design effective Type I photosensitizers for treating hypoxic tumors.

Contribution

A closed-loop hybrid system using machine learning and molecular calculations to design Type I photosensitizers is developed.

Findings

664 potential Type I PSs were identified using a support vector machine model.

Two candidates, M1 and M2, were synthesized and verified as effective Type I PSs.

M1 and M2 suppressed tumor growth through enhanced ROS generation in hypoxic conditions.

Abstract

Type I photosensitizers (PSs), which operate effectively under low‐oxygen conditions, offer a promising approach to overcome hypoxia‐associated challenges in solid tumor therapy. However, their design remains challenging due to the limited number of reported molecules with diverse structures, as well as insufficient understanding of the underlying mechanisms. Herein, a closed‐loop hybrid discovery system is developed that combines molecular excited‐state calculations with machine learning (ML) to rationally design and predict high‐performance Type I PSs for hypoxic tumor therapy. Through a support vector machine (SVM) classification model, 664 potential Type I PSs are identified from a molecular space based on donor‐acceptor (D‐A) and donor‐acceptor‐donor (D‐A‐D) structures. Among these, two candidates, M1 and M2, are synthesized and experimentally verified as Type I PSs, exhibiting aggregate‐induced enhancement of Type I reactive oxygen species (ROS) generation. Both in vitro and in vivo studies demonstrated their ability to induce intracellular Type I ROS generation and effectively suppress tumor growth. The work highlights the potential of ML in the design and prediction of Type I PSs for hypoxic tumor therapy.

The work developed a closed‐loop hybrid discovery system to rationally design and predict high‐performance Type I PSs for hypoxic tumor therapy. 664 Potential candidates are identified from a dataset through a support vector machine (SVM) classification model, and two candidates are experimentally verified as Type I PSs, which highlighted the potential ofmachine learning (ML) in the design of Type I PSs.

Figures