Aussie scientists design superbug-killing protein with AI

Antibiotic-resistant superbugs are steadily outrunning modern medicine, turning once-routine bacterial infections into life-threatening emergencies. Designing a custom protein to neutralize a specific pathogen used to demand years of trial and error in wet labs.

Now, computational biology is flipping that script.

Australian researchers have used artificial intelligence to design a brand-new, ready-to-use biological protein from scratch that kills antibiotic-resistant strains of E. coli. The breakthrough, published in Nature Communications, marks the first time an Australian research group has deployed AI to produce an active therapeutic protein.

Designing killers from scratch

The project was co-led by Dr. Rhys Grinter at the University of Melbourne Bio21 Institute and Associate Professor Gavin Knott at the Monash Biomedicine Discovery Institute, with PhD candidate Daniel Fox driving the core experimental work.

Traditionally, treating infections or cancers with proteins meant finding existing molecules in nature and repurposing them through slow rounds of rational design or directed evolution.

Instead, this team deployed deep learning to execute de novo design. By targeting the exact mechanisms pathogenic E. coli uses to hijack heme, the AI generated target-binding proteins in seconds. These synthetic molecules lock onto the bacterial target with high specificity, shutting down the pathogen without relying on pre-existing natural scaffolds.

Standing on Nobel shoulders

Rather than locking their tech away or starting from scratch, the Melbourne and Monash team built on the foundational methods pioneered by recent Nobel laureate David Baker.

Their end-to-end AI Protein Design Platform pulls in accessible, community-developed tools like Bindcraft and Chai, pairing them with in-house algorithms and laboratory validation.

According to Professor John Carroll, director of the Monash Biomedicine Discovery Institute, this setup puts Australia in the same league as the US and China, capable of rapidly screening and generating thousands of functional proteins for drug discovery, sensors, and novel biomaterials.

Why it matters

The superbug crisis is moving fast, and traditional drug pipelines are simply too slow to keep up. Demonstrating that an AI-generated protein can effectively kill resistant E. coli shows that bespoke, computationally generated biologics are moving from theoretical toys into practical countermeasures.

For researchers and biotech teams, the real story here is the power of open tooling. By chaining together publicly available deep learning suites like Chai and Bindcraft with targeted wet-lab testing, the team proved that high-impact drug discovery no longer requires billions in proprietary computing power. Agile academic teams can now punch well above their weight.

We think their pragmatic approach was brilliant: skip the hubris of building another foundation model, take the best open-source tools off the shelf, and point them straight at a high-stakes clinical bottleneck.

That said, killing bacteria in a lab dish is just step one. Synthetically designed proteins still face enormous regulatory and biological hurdles, particularly around human immune tolerance, toxicity, and in vivo stability. Generating a killer molecule takes seconds; proving it is safe for patients will still take time.