Reduced bacterial resistance antibiotics with improved microbiota tolerance in human intestinal: Molecular design and...

Antibiotic selectivity and bacterial resistance are critical global public health issues. We constructed a multi-class machine learning model to study antibiotic effects on human intestinal microbiota abundance and identified key features. Binding energies of β-lactam antibiotics with Escherichia coli PBP3 mutant protein were calculated, and a 2D-QSAR model for bacterial resistance was established. Sensitivity analysis identified key features affecting bacterial resistance. By coupling key features from the machine learning model and 2D-QSAR model, we designed ten flucloxacillin (FLU) substitutes that improved intestinal microbiota tolerance and reduced antibiotic bacterial resistance. Concurrently, the substitutes exhibited superior degradability in soil, aquatic environments, and under photolytic conditions, coupled with a reduced environmental toxicity compared to the FLU. Evaluations under combined medication revealed significant improvements in functionality and bacterial resistance for 80% of FLU substitutes, with 50% showing more than a twofold increase. Mechanistic analysis demonstrated enhanced binding to target proteins and increased biodegradability for FLU substitutes due to more concentrated surface charges. Reduced solvent hindrance and increased cell membrane permeability of FLU substitutes, mainly due to enhanced interactions with phospholipid bilayers, contributed to their functional selectivity. This study aims to address poor antibiotic selectivity and strong bacterial resistance, providing guidance for designing antibiotic substitutes.