Author ORCID Identifier

https://orcid.org/0000-0002-3195-7847

Semester

Fall

Date of Graduation

2025

Document Type

Dissertation

Degree Type

PhD

College

School of Pharmacy

Department

Pharmaceutical Sciences

Committee Chair

Bingyun Li

Committee Member

Matthew Dietz

Committee Member

Werner Geldenhuys

Committee Member

Paul Lockman

Committee Member

Yong Qian

Abstract

Antimicrobial resistance (AMR) and biofilm-associated infections are major global health concerns, affecting millions of people each year. The rise of antibiotic-resistant infections has exceeded the rate of novel therapeutics being discovered, leaving many serious infections difficult to treat. This creates additional complications in healthcare, as vulnerable patients, such as cancer patients, experience more severe and life-threatening infections due to their suppressed immune systems. These challenges demonstrate the need for novel therapeutics that not only target bacteria, including resistant strains, but also offer additional benefits, such as anticancer activity. Antimicrobial peptides (AMPs) have shown promising activity against many bacterial strains with a low tendency to promote resistance and have been further explored for their activity against various types of cancer. Although AMPs offer several advantages, their use in therapeutic applications is limited by toxicity towards mammalian cells. Traditional peptide design is often costly and time-consuming; therefore, exploring artificial intelligence (AI) to streamline this process could address some of these limitations.

This dissertation investigates the use of a large language model (LLM), ChatGPT, was investigated for designing peptides, including multi-functional peptides. Prompts were first optimized using prompt engineering peptides to generate peptides with antimicrobial and other functional properties, which was validated using machine learning predictions from publicly available tools. The role of amino acids, such as Z-amino acids was explored to reduce toxicity while maintaining antimicrobial activity. Building upon the findings of previous studies, optimized prompt and Gln-rich sequences were used as guiding factors for ChatGPT-generated peptides. Selected peptides were experimentally validated to confirm the predicted functional properties.

Our approach has shown that the informed task prompt using ChatGPT generated peptides with high predicted activity for single-, dual-, and multi-functional peptides and high accuracy. Among the Z-amino acids tested, Z-Gln exhibited antimicrobial and antibiofilm activity at concentrations that were non-toxic compared to the other amino acids. Therefore, Gln-rich constraint was incorporated into the design prompt to generate peptides with potentially lower toxicity. Three ChatGPT-designed peptides were designed: (1) antimicrobial and anticancer, (2) antimicrobial and Gln-rich, and (3) antimicrobial, anticancer, and Gln-rich were experimentally validated. Although two sequences did not have their predicted activity (2 and 3), peptide (1) demonstrated both antimicrobial and anticancer activity, highlighting both the potential and limitations of AI-generated sequences and emphasizing the need for laboratory confirmation.

Overall, this work demonstrates the potential of applying ChatGPT to peptide design through in silico validation and screening combined with experimental validation, addressing key limitations associated with traditional peptide design techniques. By leveraging ChatGPT’s user-friendly interface, this lowers the barriers commonly associated with computation peptide discovery and provides a promising approach to accelerate the development of novel peptide candidates.

Download

Available for download on Thursday, December 10, 2026

Share

COinS