How we're using AI to discover new antibiotics | Jim Collins

Episode Summary

Executive Summary: Jim Collins argues that AI and synthetic biology can dramatically accelerate drug discovery and diagnostics for COVID-19 and future pandemics. Building on a machine-learning breakthrough that found halicin, he outlines using trained models to search vast chemical spaces for antivirals, plus engineered diagnostics and a BCG-based vaccine candidate.

Main Topics: AI-driven drug discovery (Priority: 5/5): Collins explains how machine learning can replace slow, trial-and-error screening by searching enormous molecular spaces for promising therapeutics, including antivirals for SARS-CoV-2. Halicin as proof of concept (Priority: 5/5): He uses the discovery of halicin to show that AI can identify novel antibiotics that human experts would likely miss, and that resistance may emerge more slowly. Scaling antibiotic discovery (Priority: 4/5): The Audacious Project funding will expand the Antibiotics AI effort to discover seven new antibiotic classes against seven deadly bacterial pathogens over seven years. Synthetic biology diagnostics (Priority: 4/5): The talk describes freeze-dried RNA sensor systems embedded in paper and cloth, including a face-mask diagnostic that could detect infection through breathing. Vaccine engineering with BCG (Priority: 3/5): Collins proposes repurposing and engineering the BCG vaccine to express SARS-CoV-2 antigens as a scalable, safe vaccine platform. Science as a response to pandemics (Priority: 4/5): The broader message is that combining AI, synthetic biology, and rapid experimentation can help humanity outpace superbugs and emerging viruses.

Key Arguments: Machine learning can search the space of essentially all synthesizable molecules, making drug discovery far more efficient than testing compounds one by one. The same AI platform built for antibiotic discovery can be adapted to find antiviral compounds against SARS-CoV-2. Halicin validates the approach because it was discovered by a model trained on about 2,500 compounds and was not structurally similar to known antibiotics. Halicin appears unusually hard for bacteria to resist, strengthening the case for AI-guided discovery of new drug classes. Synthetic biology can create low-cost, rapid diagnostics by freeze-drying cellular machinery and RNA sensors into paper or cloth. A face mask can be turned into a diagnostic device that detects viral infection and signals results with fluorescence. BCG is a promising vaccine scaffold because it is scalable and has a strong safety profile, and can be engineered to present SARS-CoV-2 antigens.

Data Points: Compounds tested in pilot project: ~2,500 - Training set used to teach the model antibacterial activity Drug repurposing library size: Several thousand molecules - Library screened to identify candidates unlike existing antibiotics In-silico library size: Over 1 billion molecules - Search space for potential antiviral compounds New antibiotic classes discovered in last three decades: 0 - Context for the urgency of antibiotic discovery Resistance to ciprofloxacin after lab exposure: After 1 day - Comparison showing rapid resistance development Resistance to halicin after lab exposure: None after 1 day; none after 30 days - Evidence of low resistance emergence Antibiotic pathogens targeted by Audacious Project funding: 7 - Goal for new antibiotic classes against deadly bacterial pathogens Project timeline: 7 years - Timeframe for the scaled antibiotic discovery effort Diagnostic turnaround time: 1 or 2 hours - Estimated time for mask-based COVID-19 diagnosis Vaccine platform age: Almost a century - BCG has been used against TB for nearly 100 years

Pivotal Quotes: "instead of looking for a needle in a haystack, we can use the giant magnet of computing power to find many needles in multiple haystacks simultaneously" — Jim Collins: Describing the advantage of machine learning in drug discovery "In the case of Cifro, after just one day, we saw resistance. In the case of halicin, after one day, we didn't see any resistance." — Jim Collins: Explaining halicin’s resistance profile in lab tests "the simple act of breathing, along with the water vapor that comes with it, can activate the test" — Jim Collins: Describing how the mask-based diagnostic would work

Implications: If successful, these tools could speed up discovery of antibiotics, antivirals, diagnostics, and vaccines, reducing pandemic response time and helping counter antibiotic resistance with scalable, low-cost technologies.