Topology-Aware Generation and Activity-Based Filtering: A Computational-Experimental Framework for Data-Scarce...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Antimicrobial Drug Discovery

Background:

• Quaternary ammonium compounds (QACs) are vital antimicrobials facing rising bacterial resistance.
• Developing novel QACs is challenging due to limited structure-activity data and generation methods.
• Existing methods struggle to efficiently identify promising new QAC candidates.

Purpose of the Study:

• To compare two AI-driven computational workflows for accelerating QAC discovery under data-limited conditions.
• To evaluate the impact of predictive computational filtering on the quality of novel QAC candidates.
• To identify novel QACs with validated antimicrobial activity against key bacterial pathogens.

Main Methods:

• Utilized a topology-aware variational autoencoder for generating novel QAC structures.
• Workflow 1: Direct expert evaluation of generated QACs.
• Workflow 2: Computational filtering of candidates for predicted antimicrobial activity before expert evaluation.
• Experimental validation of minimum inhibitory concentrations (MICs) for selected compounds.

Main Results:

• Workflow 2 significantly improved candidate quality: synthesis-worthy compounds increased from 9% to 38%, and invalid outputs decreased from 21% to 0%.
• Expert evaluation time constraints were met more effectively with computational prefiltering.
• Eleven novel QACs with validated MICs ranging from 1-32 μM against four bacterial pathogens were identified.

Conclusions:

• AI-guided generation coupled with computational prefiltering effectively navigates data-scarce chemical spaces for drug discovery.
• This approach enhances the efficiency and success rate of identifying novel antimicrobial agents.
• The developed workflows provide a systematic strategy for discovering QACs with potent bioactivity, addressing the challenge of antimicrobial resistance.