Identifying Structure-Activity Relationships for Cyanine-Derived Antibiotics Using Machine Learning and Commercial...

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Antimicrobial resistance is a growing global health threat, necessitating novel antibiotic development.
• Understanding structure-activity relationships (SAR) is key for designing effective antibiotic scaffolds.
• The vast chemical space of potential antibiotics hinders mechanism of action elucidation.

Purpose of the Study:

• To elucidate the SAR of novel cyanine molecules as potential antibiotics.
• To compare the efficacy of traditional machine learning (ML) and large language models (LLMs) in SAR analysis.
• To identify key structural features contributing to the antibacterial activity of cyanine compounds.

Main Methods:

• Synthesis of a novel set of cyanine molecules.
• Analysis of antibacterial activity data using traditional ML classifiers.
• Application of commercially available LLMs (Grok-3 Think, ChatGPT o1) for SAR analysis.
• Comparative evaluation of ML and LLM performance in predicting antibacterial properties.

Main Results:

• Both ML and LLM approaches identified key SAR features for cyanine antibiotics.
• Certain LLMs, specifically Grok-3 Think and ChatGPT o1, demonstrated superior performance compared to traditional ML classifiers.
• Positive charge and lipophilicity were highlighted as critical physicochemical properties for potent cyanine-based antibacterial agents.

Conclusions:

• LLMs show promise as powerful tools for accelerating antibiotic drug discovery by elucidating SAR.
• The study identified specific structural features (positive charge, lipophilicity) crucial for developing effective cyanine antibiotics.
• Integrating advanced computational methods like LLMs can overcome challenges in understanding complex SAR for novel antimicrobial agents.