🧬 A thank you, and an open invitation. Over the past few weeks we shared our AMR landscape report and five proposals built on it. Huge thanks to the 130+ researchers, 60+ workshop experts, and working groups who made it happen. The proposals are the starting line, not the finish. The real work of building these datasets and models is still ahead, and we'd love help: data, collaboration, funding. Here's to seeing it get done. 📄 hubs.la/Q04sJ50g0 📬 [email protected] 🌐 hubs.la/Q04sJ0kx0 #AMR #OpenScience #BioAI

🧬 Proposal spotlight: predicting resistance before it emerges in a patient. When a patient turns from susceptible to resistant mid-treatment, is it new mutation or reinfection? We'd build a prospective, multi-site LMIC cohort (an ACORN extension) with paired WGS to predict it, and tell the two apart. The gap: almost no ML-for-AMR work uses LMIC data. 📄 hubs.la/Q04sJ0gR0 📬 [email protected] #AMR #BioAI #GlobalHealth

🧬 Proposal spotlight: flagging antibiotic liver toxicity before synthesis. Toxicity models trained on typical drugs stumble on antibiotics. We'd build the first matched antibiotic dataset across liver cell lines and organoids, then predict hepatotoxicity and its mechanism from structure. The gap: antibiotics are barely present in existing toxicity data. 📄 hubs.la/Q04sHYks0 📬 [email protected] #AMR #DrugDiscovery #BioAI

🧬 Proposal spotlight: personalized antibiotic dosing for kids. Start by extending KidsCalc, a validated vancomycin dosing tool, beyond Australia. Then build a model predicting outcomes, toxicity, and resistance emergence, so dosing adapts to each child. The gap: the data pieces exist but aren't linked to outcomes. 📄 hubs.la/Q04sHNWH0 📬 [email protected] #AMR #BioAI #GlobalHealth

🧬 Proposal spotlight: getting drugs into Gram-negatives. Their outer membrane is the problem: will a compound get in, and will it stay in? We'd build an LC-MS/MS accumulation dataset across E. coli, A. baumannii, and P. aeruginosa, using active learning to pick the most informative compounds. The gap: almost no direct accumulation data for two of the three. 📄 hubs.la/Q04sHR820 📬 [email protected] #AMR #DrugDiscovery #BioAI

🧬 Proposal spotlight: predicting quantitative MIC straight from a genome. Resistance is shaped by genes, mutations, copy number, and epistasis. We'd build species-specific MIC-from-WGS models across ~10 WHO priority pathogens, on globally diverse isolates so predictions hold up beyond where the data came from. The bottleneck: data that's sizable but geographically narrow. 📄 hubs.la/Q04sHH5z0 📬 [email protected] #AMR #OpenScience #BioAI

Dive into the proposals. Looking for feedback and collaborators. 📄 zenodo.org/communities/align 📬 [email protected] #AMR #OpenScience #BioAI

Emergence of resistance in a patient: prospective LMIC cohort to predict treatment-emergent resistance.

Antibiotic hepatotoxicity: screen 10,000 compounds across five liver injury mechanisms.

Effective antibiotic dosing in pediatrics: extend KidsCalc beyond Australia, predict treatment failure & toxicity.

Compound entry & efflux in Gram-negatives: measure intracellular accumulation with active learning.

Quantitative MIC from WGS: build a globally representative dataset to test cross-species prediction.

From priorities to proposals. Our AMR landscape report mapped where AI could help most. Working groups have now turned five of those priorities into concrete proposals 🧵

What comes next: five focused proposals that actually build these datasets and models. But they need funders, collaborators, and data contributors to move forward. If your work aligns with these priorities, let's connect. 📄 Zenodo doi.org/10.5281/zenodo.21414… 📬 [email protected] #AMR #OpenScience #BioAI

The community prioritized 10 areas where AI and good datasets could move the needle: from quantitative MIC prediction and Gram-negative permeability to dosing, resistance emergence, and One Health burden.

🧬 A closer look at our AMR landscape report (with @GoogleDeepMind). Around 5M deaths were associated with AMR in 2019. Up to 10M a year is projected by 2050 without action. This is urgent. And it's solvable.

This answers the first call to action from the AI for AMR report, from @GoogleDeepMind and the Fleming Initiative: consensus on the highest-priority "learnable" problems. The proposals come next. 📄 Zenodo 🌐 alignbio.org #OpenScience #BioAI #AMR

So we mapped what the community sees as most urgent: 🔹 130+ experts, 20+ countries 🔹 60+ experts, 6 workshops, 3 world regions 🔹 10 problem areas prioritized 🔹 10 working groups now building proposals

AMR is full of problems AI is well suited to (complex, multimodal, high-dimensional), but the data is fragmented and rarely comparable across labs. What would it take to change that?

🧬 New from Align, with support from @GoogleDeepMind: a community landscape report on where AI can make the biggest dent in antimicrobial resistance. How can AI help? We asked the field one question: what data would be most useful for AI/ML models tackling AMR? 🧵