AI is changing not only how we search for antibiotics, but increasingly how we design them. Our latest piece is out in @CellBiomat @CellPressNews. We examine how AI is reshaping antibiotic discovery, from machine learning for screening and optimization to protein language models, graph neural networks, and generative approaches for de novo antimicrobial design. With antimicrobial resistance continuing to outpace antibiotic discovery, the opportunity is bigger than searching faster. These tools can help us navigate chemical and biological space more systematically, connect computation with experiment, and find molecules we might otherwise miss. A wonderful collaboration with Jing Xu, Chen Li, Xiaoyu Wang, Anton Y. Peleg, @lifuyi1991, @supercs08, and colleagues across @MonashUni and @Penn, including @CBE_Penn, @PennBioeng, @PennEngineers, @PennMedicine, @PennChemistry, @PennMicro, @PennPsych, and @PennSAS. Read the paper: doi.org/10.1016/j.celbio.202…

Mitochondria are descendants of bacteria. Could their own genomes still encode molecules with antimicrobial activity? In our new work, we used AI to explore over 2,000 ancient human mitochondrial genomes and identify previously hidden peptide sequences with antibacterial potential. We call these molecules mitochondrins. Experimental testing showed that several mitochondrins are active against clinically relevant bacteria, including drug-resistant strains. They display distinct mechanisms of action, and a representative peptide reduced A. baumannii burden in a mouse infection model. Intriguingly, several active sequences are perfectly conserved in modern human mitochondrial DNA. The broader question is perhaps the most interesting: could the mitochondrial genome itself represent a previously underexplored layer of human innate immunity? We do not yet know whether mitochondrins are naturally produced or deployed during infection. Establishing that will require further work. But our results provide a biochemical basis for testing the idea that mitochondrial genomes may contribute directly to host defense. More broadly, this study reflects a direction we are deeply interested in: using AI to interrogate biological information across evolutionary time—not simply to catalog what is there, but to uncover function. The code of life contains far more functional information than we currently understand. Machines can help us begin to read it. Great team work @mdt_torres, Abdullah Ali, Hyun-Su Lee, Fangping Wan, @Penn, @CBE_Penn, @PennBioeng, @PennEngineers, @PennMedicine, @PennChemistry, @PennMicro, @PennPsych, @PennSAS! Preprint link: biorxiv.org/content/10.64898…