An AI-Designed Cancer Therapy Just Entered a Child's Dose-Escalation Trial — the Hard Part Starts Now
Healx dosed its first patient in a Phase 1/2 sarcoma trial, and the bottleneck moves from discovery to trial operations. Plus: Chai Discovery's $400M Series C at $3.8B, Insilico takes generative AI onto the factory floor, Doudna's lab builds a nuclease evolution never made, and The Onion Desk.
Healx announced first-patient dosing in a Phase 1/2 trial of HLX-4310, an oral combination pairing domatinostat (a Class I-selective HDAC inhibitor) with sirolimus (an mTOR inhibitor), targeting metastatic recurrence in sarcomas — particularly osteosarcoma, where survival after metastasis sits below 30% and the patients are adolescents and young adults. What AI contributed was the combination rationale, not a new molecule. That distinction matters, because platforms now reliably push candidates into first-in-human testing while the machinery around them — dose-escalation methodology, enrollment models, safety monitoring — was designed for a slower discovery cadence.
Lead — an AI-designed therapy reaches a child's dose-escalation trial
ClinicalHealx dosed its first patient in a Phase 1/2 trial of HLX-4310, combining domatinostat with sirolimus against metastatic recurrence in sarcomas — osteosarcoma above all, a disease of adolescents and young adults with survival below 30% after metastasis. The candidate represents AI's contribution to combination rationale rather than novel molecule design, a narrower and more honest claim than most 'AI drug' headlines carry. The revealing number is the shape of the funnel: industry data shows AI-discovered candidates clearing Phase 1 at 80–90% against a historical ~50%, then reverting to the standard ~40% at Phase 2. Improving the odds of surviving a safety readout is not yet evidence of improving the odds of showing efficacy. What follows isn't a scientific bottleneck but an operational one — enrollment timelines, protocol adaptation, and regulatory alignment across pediatric and adult populations, in a disease where the eligible population is small and the trial infrastructure was built for slower discovery.
Chai Discovery raises $400M, betting pharma will buy models instead of molecules
FundingChai closed a $400M Series C at a $3.8B valuation, pushing total funding past $600M. Index Ventures led; OpenAI and returning investors cited deployment at Lilly, Pfizer, and Novartis. Chai's platform generates antibodies and proteins computationally, and Chai-2 reported double-digit lab success rates for zero-shot antibody design — roughly a 100x improvement over prior methods. The valuation prices in the belief that 'designed, not discovered' survives its first clinical setbacks.
Insilico takes generative AI off the discovery bench and onto the factory floor
DealsInsilico and Bora Pharmaceuticals launched a strategic alliance deploying generative AI into pharmaceutical manufacturing, with potential value exceeding $2.5B — a deliberate move downstream from discovery into process and manufacturing optimization, where regulatory comfort is higher and ROI is easier to attribute. Insilico now claims 13 partnerships among the top-20 global pharmas, including Lilly (~$2.75B), SK Biopharmaceuticals ($2.5B), and CMS ($165M), with 31 preclinical candidates nominated since 2021 and 13 advanced to IND clearance.
A Nobel laureate enters AI protein design — and builds an enzyme evolution never made
PapersJennifer Doudna's lab published a platform in Science for generating novel gene-editing enzymes, designing a never-before-seen protein that binds and cuts DNA. The team built non-natural TnpB nucleases operating outside evolved sequence space, with demonstrated activity in human, plant, and bacterial cells. It moves AI protein design past binders and into functional nucleases — the core machinery of gene editing, and computationally designed tools for gene therapy rather than computationally designed decorations on them.
BMS and NVIDIA add a second DGX SuperPOD
InfrastructureBristol Myers Squibb deployed a second DGX SuperPOD on DGX Vera Rubin NVL72 systems claiming 10x performance gains, opening it and the BioNeMo Agent Toolkit to its scientists — the emerging advantage lies in compute plus proprietary data rather than any individual model.
Insilico and China Medical System sign a ~$177M CNS deal
DealsA ~$177M CNS collaboration targeting novel mechanisms identified through PandaOmics — Insilico's second announcement of the week, and a reminder that its deal cadence now runs across both discovery and manufacturing.
argenx takes early access to Chai's platform
Dealsargenx gained early access to Chai's platform for computationally designed antibodies — Chai's fourth disclosed pharma partnership in seven months, and part of what the $3.8B valuation is buying.
The Onion Desk
A top-ten drugmaker achieved 'zero AI incidents' by deploying a fourteen-layer review governance structure that has shipped no models since March. Biotech seed founders located the industry's $9.1B in venture funding and confirmed it exists — entirely inside mega-rounds and late-stage companies. And a multi-agent system designed a flawless Phase 3 protocol; the simulated patients declined enrollment. Full satirical dispatch on Substack.