This week's paper 👇 📄 "SMILES-Based Deep Learning Enables Discovery of Chemically Modified Ultrashort Antimicrobial Peptides" by Hu et al., Journal of Medicinal Chemistry, October 22, 2026 🧪 What was studied: Antimicrobial peptides are short chains of amino acids that can attack bacteria. Most computer models for designing them only understand the 20 standard amino acids. This team built a model (called SMAMP) that reads a peptide's full chemical structure, so it can also handle modified or unusual building blocks. ⚙️ How: They trained the model to predict how strongly peptides block E. coli growth in the lab. Then they used it to design 19 very short peptides, made them, and tested them in lab dishes. The two best candidates were also tested in wax moth larvae and in mice. 📊 What they reported: 13 of the 19 designed peptides showed antibacterial activity in lab tests, and the model's predictions lined up fairly well with the lab measurements. The two lead peptides showed low toxicity to mammalian cells in the dish, and in their tests bacteria developed resistance to them more slowly than to standard antibiotics. Their data suggest the peptides work by damaging the bacterial membrane. ⚠️ Keep in mind: This was cell and animal research only. No humans were studied. It's a small set (19 peptides, only 2 taken into animal tests), and early lab results often don't hold up later. The abstract doesn't list the authors' limitations and the full text is paywalled, so check the paper for their own caveats. Early findings, and observational ones especially, don't prove cause and effect. These are brand-new experimental compounds, not anything we sell. 🔗 Source: https://pubmed.ncbi.nlm.nih.gov/42911719/ 💬 Your turn: Do you think AI-designed peptides will speed up research, or will lab testing always be the bottleneck? For Research Use Only. Not for human or animal consumption.