Quantitative analysis of antimicrobial peptide efficacy using AFM-based nanomechanics and machine learning models

Quantitative evaluation of antimicrobial peptide (AMP) activity is important for identifying promising peptide candidates and supporting their development. Here, we propose a new method for predicting and evaluating AMP activity by integrating atomic force microscopy (AFM) imaging, quantitative nanomechanical measurements, and machine learning. Using FC8-A12L (A12L) against Escherichia coli as a model system, AFM Quantitative Imaging simultaneously characterized bacterial height, adhesion force, and Young's modulus at different peptide concentrations. A dual-channel threshold-intersection strategy combining height and Young's modulus maps was developed to identify bacterial regions and automatically extract spatially matched measurements of all three parameters. These measurements revealed changes in bacterial morphology, a biphasic variation in adhesion, and progressive mechanical softening with increasing A12L concentration, providing complementary biophysical information for assessing antibacterial effects. Machine learning integrated the extracted parameters to establish quantitative relationships between bacterial biophysical properties and A12L treatment levels. The multilayer perceptron showed the best predictive performance in the comparison of untreated and 25 μg/mL-treated bacteria. Prediction across 0, 12.5, 25, and 50 μg/mL achieved a test R² of 0.99897 and a root mean squared error of 0.59210 μg/mL. By connecting AFM measurements with automated feature extraction and predictive modeling, this study provides a new approach to quantitative evaluation of AMP activity and a complementary tool for assessing peptide efficacy.