Boltz

New in Boltz Lab: interactive small-molecule clustering, a smarter structure viewer, and more! We've shipped a series of updates to Boltz Lab that make it easier to analyze results, work with structures, and manage your projects. Among the additions is a new analysis experience for small-molecule screens: interactive UMAP and t-SNE plots that let you visualize how molecules cluster, explore the diversity of your results, and identify groups worth further investigation. We've also rolled out: - Import custom IDs, columns, and metadata with your virtual libraries - Toggle sidechains and apply custom color profiles in the structure viewer - Bulk export structures for offline analysis - Support for reference structures within projects, so you can evaluate designs against known structural data - Add notes to sandbox predictions to capture hypotheses and observations These updates are available now, with more on the way. Many of them originated directly from user feedback! Interested in trying Boltz Lab? Request access at https://lnkd.in/eqrmU9pU For partnership inquiries, contact partnerships@boltz.bio

Happy to share that Boltz is a founding member of a new data consortium organised by A-Alpha Bio, alongside GSK, Cradle, and Dyno Therapeutics! Training data is one of the important ingredients in building generalisable biomolecular foundation models. Consortia are a great way of generating data in a way that provides the most value for partners and society. Thank you to A-Alpha Bio and our fellow partners for making this possible. Excited for what everyone will build on this new data! More on Andrew's article: https://lnkd.in/ewUKPaHs

Thank you Andrew Dunn and Endpoints News for covering the launch of our Atlas Data Consortium. At A-Alpha Bio, we are eliminating the data bottleneck facing AI-driven antibody engineering. Even with our AlphaSeq platform, high-throughput experimental data generation is not enough to train generalizable models. We are demonstrating how the cost of foundational affinity and structure data can be shared across the industry to achieve the data quantity and diversity needed for model generalizability. As long as data remains siloed, no single organization will close the gap alone. The Atlas Consortium is a first-of-its-kind structure designed to change that. Consortium members and A-Alpha cooperate to design and generate new experimental data prospectively, then pool it. We are proud to launch with an outstanding group of founding members, including GSK, Boltz, Cradle, and Dyno Therapeutics. Building a cooperative model requires organizations willing to commit to the vision before the flywheel has had a chance to turn. We are grateful for their trust and partnership. With data at the scale this model enables, generalizable AI for de novo design and zero-shot optimization is within reach for any organization working to advance protein therapeutics. Built around pre-competitive datasets with no proprietary sequences or targets, and with dedicated pathways available outside the Consortium for proprietary data needs, the Atlas Consortium is proof that the industry can come together to solve shared data challenges, and a way to ensure that powerful protein AI tools are accessible to any organization, regardless of the size of their internal data infrastructure. Learn more about the Atlas Consortium: https://lnkd.in/gQ9n-SqU Read the story: https://lnkd.in/gQYmhtvB

By accessing Boltz-2.1 via its API, Vecura brings structure-confidence screening directly into an orchestrated de novo protein design workflow. In this case study, #RFdiffusion generated candidate backbones, #LigandMPNN designed sequences, and Boltz-2.1 folded each sequence with ten sampled models to produce confidence signals for ranking and filtering. Across nine designed sequences targeting the ATP pocket of the LRRK2 kinase domain, the workflow returned full 3D structures together with confidence metrics that helped identify which candidates were more credible and which required further iteration. The scores were used for relative triage, rather than as proof that any candidate was definitively well-folded. This is especially important for early-stage de novo designs, where confidence values should be interpreted carefully. For research teams, the broader advantage is flexibility. Vecura can connect pre-integrated tools with models accessed through external APIs, allowing teams to build complete workflows without moving files between services or maintaining separate environments for every model. Bring your own design pipeline to Vecura and use Boltz-2.1 confidence screening to fold, rank, and filter a full generation of designs before the next iteration. ▶️ Read the full workflow and results: https://lnkd.in/ghNrA9yt ▶️ Explore Vecura: vecura.com ▶️ Start your research: app.vecura.com #ProteinDesign #DeNovoProteinDesign #StructuralBiology #AIforScience #ComputationalBiology #Boltz #Vecura

Astellas Pharma just publicly shared how, using Boltz's models on a membrane protein target, they were able to find validated molecules performing on-par in-vitro to a clinical compound while reducing the number of assays needed by 90% and shortening the timelines by 70%! This week Jeremy shared the stage with NVIDIA Healthcare and Astellas Pharma in Tokyo and Osaka sharing how our models are accelerating drug discovery. Boltz models are successfully reducing the experimental bottleneck in real-world drug discovery campaigns across both small-molecules and biologics. You can start using the Boltz API today or reach out to me if you'd like to explore our full enterprise platform! Reach out to me at contact@boltz.bio to access our enterprise platform Read more about Astellas's announcement: https://lnkd.in/gNKBAafg  Read more about the Boltz API and the experimental validation of our latest models: https://lnkd.in/gZSwW7hm

Excited to see this preprint, congratulations to everyone at Boltz. Looking forward to continuing our collaboration!

A bit of a belated announcement, but I'd like to share that I have joined Boltz as a Member of Technical Staff! A few years ago, my experience at Meta's FAIR Chemistry team ignited my passion for using machine learning to advance scientific discovery. I became convinced that the intersection of AI and chemistry holds immense potential to solve some of the world's most pressing challenges, and I joined a PhD program to build the depth needed to contribute to that frontier. I'm extremely grateful to my advisor, Pan Li, for his mentorship along the way. Boltz sits right at that frontier. Its mission is genuinely inspiring, and our open-source models are in the hands of scientists around the world, powering real research today. I truly believe that the work we are doing will have a meaningful impact on drug discovery, and I am honored and excited to be part of it. It has been a joy to work alongside such an incredible team: Gabriele, Jeremy, Saro, Noah, Yunguan, Alvaro, Talip, Hannes, Charlie, Avene, Geoffrey, Jack, Armando, Nicola, Francesco, Simone, Demitri, Luis, Zach, Jose, Luca, Thomas, Matthew, and Rita. We are actively looking for talented, motivated researchers and engineers. If that sounds like you, please reach out to me or anyone at Boltz!