Identifying Structure–Activity Relationships for Cyanine-Derived Antibiotics Using Machine Learning and Commercial Large Language Models

Understanding the structure–activity
relationship (SAR)
of antibiotic scaffolds is crucial for the development of antibiotics
to counter the growing crisis of antimicrobial resistant bacteria.
However, an overwhelming space of structural features impairs a comprehensive
understanding of the mechanism of action for potential antibiotic
candidates. In this study, antibacterial data of a set of newly synthesized
cyanine molecules are analyzed with both traditional machine learning
(ML) and commercially available large language models (LLMs) to elucidate
the SAR. Some LLMs, particularly Grok-3 Think and ChatGPT o1, outperform
the traditional ML classifiers, and both approaches highlight positive
charges and lipophilicity as key properties for effective cyanine
antibiotics.