XtalPi

Plastic recycling has stayed under 10% for decades, partly because durable plastics are hard to break down cleanly. New plastics developed by XtalPi-incubated FutureBio’s founders recover 93-96% of their triketone building blocks in 5 M HCl over 24-72 h at just 20-60 °C. Computational screening of 144 candidate monomers identified solvation free energy as the main factor controlling depolymerization temperature. The team then engineered modular PKS enzymes to biosynthesize structurally varied building blocks with defined stereochemistry. Those structural differences set both material performance and recycling conditions. The polymers show tunable glass-transition temperatures (53-98 °C), mechanical properties, and solvent resistance, and can be dismantled at different temperatures within the 20-60 °C window. This enables staged recovery of mixtures of these PDK materials, with 93-96% monomer recovery demonstrated in laboratory tests. Commercial-scale modeling of the biomanufacturing route from renewable feedstocks, such as agricultural waste, projects monomer costs well below the petrochemical alternative (~$1.34/kg versus ~$10/kg for dimedone) and a lower carbon footprint. FutureBio has expanded its pipeline to additional bio-based monomers and circular materials, with PKS titers already at pilot-testable levels, and is in technical discussions with partners across electronics, automotive, home products, and personal care. #AIforScience #Bioplastics #CircularMaterials

99% yield and 92% ee at room temperature, with just 5 mol% catalyst. Getting high reactivity and high selectivity together is the challenge. XtalPi and SUSTech used computational design to achieve both. We combined DFT calculations and noncovalent interaction analysis to guide the development of chiral antimony catalysts. This helped us identify structural changes that strengthen reactivity while favoring the desired enantiomer, then make and test those designs experimentally. Our optimized catalyst, CAC-10, delivered the performance above. Across the study, we demonstrated the chemistry with 33 epoxide-aniline combinations, producing β-amino alcohols, building blocks found in pharmaceuticals and natural products. Published in JACS, this work shows how computation can turn an understanding of chemical reactivity into the ability to design for it. Moving from trial-and-error screening toward predictive design offers a faster, less costly route to the chiral catalysts used in pharmaceutical synthesis. #AIforScience #AsymmetricCatalysis #JACS

Six days to #SWITCH2026! XtalPi’s co-founder Dr. Lipeng Lai and chief scientific officer Dr. Peiyu Zhang will take the stage to discuss drug discovery, TechBio, and materials AI. We’re also hosting our #AIforScience partner event on Oct 28, with guests from NVIDIA, L.E.K. Consulting, and INFevo, where we’ll explore how AI models, scientific agents, and self-driving labs are advancing drug discovery. Join us in Singapore. Partner event registration👇

Meet lcms2yield, XtalPi’s state-of-the-art model that predicts UV correction factors up to 1,000,000x faster than quantum chemistry, now open source. Already running in our automated labs, it predicts full-spectrum UV correction factors from a SMILES string in milliseconds. These predictions, together with LC-UV data and an internal standard, let us estimate reaction yields without purifying product standards or building product-specific calibration curves. On 178 real reactions (Suzuki, amide coupling, Buchwald-Hartwig, SN2): • 4.74 percentage-point mean absolute error vs traditional calibration curves • 59% within 5 percentage points, 89% within 10 Pretrained on QC calculations for 22,609 compounds, fine-tuned on 1,845 experimental measurements. Published in JACS Au. Links below 👇 What’s one task in your lab you wish you never had to do manually again? #AIforScience #OpenSource

Our first siRNA PCC is here, developed internally in just 9 months with Kodexia™. The IgAN candidate’s IC50 was tens of times lower than clinical-stage references in vitro. In nonhuman primates, one dose suppressed the target protein for months, longer than either reference.🧵