HyperLab reports rapid SBDD workflow
HyperLab unveiled an AI‑driven structure‑based drug design workflow that integrates pose prediction, virtual screening at scale and ADME/T evaluation in a single interface, claiming a 9% hit rate from a 24‑hour screen. The platform highlights the push to compress discovery cycles using high‑throughput virtual screening and integrated design tools. (x.com)
Drug hunters often start with a protein’s 3D shape, then test which molecules might fit it like a key in a lock. HyperLab says it now bundles that search, ranking and safety screening into one web workflow. (biorxiv.org) In a preprint posted on bioRxiv, HyperLab’s team at HITS in Seoul said the platform predicts how a small molecule sits in a protein pocket, screens chemical libraries and estimates absorption, distribution, metabolism, excretion and toxicity before lab work begins. The company markets HyperLab as a web-based system for “binding, ADME/T, design, and screening” in one place. (biorxiv.org, hyperlab.ai) The paper says HyperLab can search libraries ranging from 1 million to 7 trillion compounds, and that one internal screening run finished in 24 hours. In that study, the team reported five experimentally validated compounds with half-maximal inhibitory concentration values from 70 to 600 nanomolar. (biorxiv.org) Drug discovery teams use this kind of structure-based design because wet-lab screening is slow and expensive, and many groups still stitch together separate tools for docking, property prediction and molecule design. A 2019 review in *Molecules* described virtual screening and de novo design as core parts of the field, while a 2024 review in *Drug Discovery Today* said companies are increasingly trying to integrate artificial intelligence across discovery and development. (ncbi.nlm.nih.gov, sciencedirect.com) The company’s pitch is speed with fewer handoffs. HyperLab’s site says researchers can move from hit discovery to absorption, distribution, metabolism, excretion and toxicity prediction without leaving the platform, and the preprint says the interface is aimed at scientists without deep artificial intelligence or computational chemistry training. (hyperlab.ai, biorxiv.org) The technical claim behind the workflow is that HyperLab’s “Hyper Binding” model reached 77% accuracy on the PoseBuster v2 benchmark for binding-pose prediction. The same preprint says AlphaFold 3 reached 84% on that benchmark, but HyperLab argues its model runs faster for screening work. (biorxiv.org) HyperLab has been building toward this for several years. HITS introduced the platform publicly at the AIMECS 2023 medicinal chemistry meeting in Seoul, describing it then as a cloud software service for predicting drug-protein interactions and designing new structures. (laplacepartners.co.kr) The 9% hit-rate figure now circulating appears in HyperLab’s own product materials and blog posts, not in an independent journal paper. The bioRxiv preprint reports five validated hits from a top-ranked set, but it does not, in the excerpt available, spell out the denominator needed to verify a 9% rate from that experiment. (hyperlab.hits.ai, biorxiv.org) That leaves the story where many artificial intelligence drug-discovery claims land: a concrete workflow, a fast internal case study and early experimental data, with broader validation still to come. For labs trying to cut months from early screening, the next test is whether those numbers hold up outside HyperLab’s own stack. (biorxiv.org, hyperlab.ai)
