Artificial intelligence is rapidly reshaping medicine, but some of its most promising applications may occur long before a patient ever receives treatment.

A new study led by Houston Methodist researcher Dr. Eleftherios Mylonakis, with Dr. Fadi Shehadeh and Dr. Biswajit Mishra serving as co-first authors, demonstrates how AI can dramatically accelerate one of the most difficult challenges in infectious diseases: designing entirely new antimicrobial candidates capable of combating antibiotic-resistant bacteria.

Published in Nature Communications, the research introduces an AI-powered platform called CAMPER — short for Constraint-Driven AMP Engineering with Ranking — that successfully designed a novel antimicrobial peptide capable of targeting methicillin-resistant Staphylococcus aureus (MRSA), including the notoriously difficult “persister” cells and biofilms that often evade conventional antibiotics.

For clinicians, the findings represent more than another machine learning model. They offer a new blueprint for accelerating therapeutic discovery at a time when antimicrobial resistance continues to outpace the development of new antibiotics.

Antimicrobial peptides are short molecules found across diverse organisms — including components of the immune system and insect venoms — that destroy bacteria by disrupting their outer membranes rather than targeting the active biosynthetic pathways attacked by traditional antibiotics. Because of this unique mechanism, they are considered promising candidates against drug-resistant organisms and drug-tolerant bacterial states such as persisters. Designing effective peptides, however, has traditionally required years of trial and error.

"The challenge isn't recognizing the promise of antimicrobial peptides," says Dr. Mylonakis, chair of Houston Methodist's Charles W. Duncan Jr. Department of Medicine. "It's identifying the right ones quickly enough to make drug development practical."

CAMPER was designed to overcome these bottlenecks by combining machine learning with biologically informed constraints to rapidly prioritize peptide candidates most likely to succeed against resistant pathogens.

Unlike many AI-driven drug discovery approaches that rely solely on statistical prediction, CAMPER integrates machine learning with biophysical principles known to govern how antimicrobial peptides interact with bacterial membranes.

Specifically, the platform incorporates a biophysical scoring function that evaluates four key properties — net charge, hydrophobicity, hydrophobic moment and helicity — to determine a peptide's ability to disrupt bacterial membranes.

The system evaluates thousands of potential peptide candidates, ranking them according to both predicted antimicrobial activity and biological plausibility before selecting the strongest candidates for laboratory testing.

That strategy paid off.

The platform designed a lead candidate known as WP-CAMPER1, a 12-amino-acid peptide derived from a wasp venom (mastoparan) template, which demonstrated potent activity against MRSA in laboratory experiments and murine models. The peptide not only killed actively growing bacteria but also proved effective against biofilms and dormant persister cells — two reasons chronic bacterial infections remain so difficult to eradicate.

In preclinical testing, a 2% topical WP-CAMPER1 formulation significantly reduced bacterial burden in infected skin tissue by 2.5 log₁₀ (p < 0.0002) in a prophylactic skin infection model. Additional studies using the peptide's D-enantiomer (WP-CAMPER1-d) confirmed activity against deep-seated infections and persistent bacterial populations in a neutropenic thigh infection model.

Although the research team cautions that the current study represents proof of concept only, the implications for the technology extend well beyond a single application.

Rather than replacing laboratory science, CAMPER dramatically narrows the search space, allowing researchers to focus experimental resources on the most promising therapeutic candidates.

"The combination of computational prediction and biological validation can substantially shorten the timeline for developing future anti-infective therapies."

---

Eleftherios Mylonakis, MD, PhD

The work also reflects a broader strategy across Houston Methodist, where investigators are developing increasingly sophisticated AI models to tackle some of medicine's most pressing challenges. From identifying novel drug candidates and advancing precision medicine to improving diagnostics and uncovering new disease mechanisms, researchers are leveraging AI not simply to automate existing workflows, but to ask — and answer — scientific questions that were previously impractical because of the scale and complexity of the data involved.

As antibiotic resistance continues to rise worldwide, tools like CAMPER suggest that AI's greatest contribution to medicine may not be replacing researchers' expertise but accelerating the scientific discoveries physicians urgently need.