Interpretable Active Learning Identifies Iron-Doped Carbon Dots With High Photothermal Conversion Efficiency for Antitumor Synergistic Therapy

Abstract

Active learning (AL) is a powerful method for accelerating novel materials discovery but faces huge challenges for extracting physical meaning. Herein, we novelly apply an interpretable AL strategy to efficiently optimize the photothermal conversion efficiency (PCE) of carbon dots (CDs) in photothermal therapy (PTT). An equivalent value (SHapley Additive exPlanations equivalent value [SHAP-EV]) is proposed which explicitly quantifies the linear contributions of experimental variables to the PCE, derived from the joint SHAP values. The SHAP-EV, with an R² of 0.960 correlated to feature's joint SHAP, is integrated into the AL utility functions to enhance evaluation efficiency during optimization. Using this approach, we successfully synthesized iron-doped CDs (Fe-CDs) with PCE exceeding 78.7% after only 16 experimental trials over four iterations. This achievement significantly advances the previously low PCE values typically reported for CDs. Furthermore, Fe-CDs demonstrated multienzyme-like activities, which could respond to the tumor microenvironment (TME). In vitro and in vivo experiments demonstrate that Fe-CDs could enhance ferroptosis through synergistic PTT and chemodynamic therapy (CDT), thereby achieving remarkable antitumor efficacy. Our interpretable AL strategy offers new insights for accelerating bio-functional materials development in antitumor treatments.