Data-Driven Structure–Function Modeling of Porous Carbon Materials for Antimicrobial Performance Prediction

Abstract

Carbon porous materials have been getting increasing interests as antimicrobial due to high surface area, adjustable pore structure, and flexibility of surface chemistry. However, the antimicrobial action is determined by multifaceted and highly nonlinear interaction between structural, physicochemical, and surface functional properties, complexifying and rendering inefficient and unreliable conventional trial-and-error optimization of material. This work presents the framework of a data-driven structurefunction modelling that predicts and analyses the antimicrobial activity of porous carbon materials through the use of machine learning methods. An experimental database of porous carbon systems was assembled, key descriptors summarising these systems were incorporated which included: BET surface area, pore size distribution, micro-mesopore ratio, surface functional group density, zeta potential, hydrophobicity, and synthesis parameters. Various kinds of supervised regression models, such as random forest regression, gradient boosting regression and feedforward neural networks were used to map these descriptors to quantitative antimicrobial performance measurements, such as bacterial log-reduction and minimal-inhibitory concentration. The cross-validation and standard statistical metrics were strictly used to assess model performance. To achieve physical interpretability, explainable artificial intelligence techniques were used to determine the prevailing material properties and their synergistic interactions in determining the behaviour of antimicrobials. The findings indicate that surface chemistry and electrostatic interactions hold more important criticality than surface area and pore architecture is the most critical one and modulator of microbial accessibility and contact efficiency. Through the proposed framework, it is possible to screen virtual antimicrobial porous carbons in a few hours and rational design it, considerably decreasing expenses of the experiment and time-to-development. The paper has described the potential of data-driven materials informatics to speed up the discovery and optimization of novel antimicrobial carbon-based materials of the next generation.