AI in Healthcare

Today's Stories

1. 1

   Insilico starts Phase III trial for AI-discovered lung-fibrosis drug

   Insilico Medicine initiated a Phase III clinical trial for Rentosertib, an oral small-molecule inhibitor targeting TNIK for idiopathic pulmonary fibrosis (IPF). The drug was discovered using Insilico's AI platforms—PandaOmics identified TNIK as a high-priority fibrosis target, and Chemistry42 designed the molecule. The Phase IIa trial showed manageable safety and tolerability, with the 60 mg once-daily arm demonstrating mean forced vital capacity improvement of +98.4 mL at 12 weeks. IPF is a progressive lung-scarring disease with median survival commonly reported at approximately two to four years after diagnosis, and current approved antifibrotic therapies can only slow progression but do not reverse the disease. Rentosertib represents a potentially first-in-class medicine whose target was identified by AI, whose chemical structure was designed by generative AI, and whose development is aimed at this severe age-related disease with high unmet medical need. This marks a major late-stage milestone for AI-driven drug discovery moving from research into late-stage clinical validation.

   The Phase III trial is expected to enroll 320 patients with IPF and is designed to systematically evaluate efficacy and safety of once-daily Rentosertib administered over 52 weeks. The trial will be led by Professor Zuojun Xu of Peking Union Medical College Hospital as Leading Principal Investigator, with Academician Nanshan Zhong and President Chang Chen as Co-Leading Principal Investigators.

2. 2

   When the sovereign AI diagnosis goes prime time

   When the sovereign AI diagnosis goes prime time

3. 3

   Pharma Giants Back AI Drug Discovery with Billions as Clinical Proof Lags

   Isomorphic Labs raised $2.1 billion（約3400億円） in May led by Thrive Capital and secured major partnerships with Novartis, Eli Lilly, and Johnson & Johnson. The company's IsoDD platform predicts protein-ligand interactions and identifies cryptic binding pockets to expand the druggable landscape. Separately, Genesis Molecular AI and Incyte announced an expanded collaboration worth potentially over $1 billion（約1600億円）, while Chai Discovery licensed its Chai-3 antibody design model to Pfizer, and Inceptive partnered with Alnylam Pharmaceuticals in a deal worth up to $2 billion（約3200億円） with $30 million（約48億円） upfront. The investments reflect conviction in AI platforms and proprietary datasets as the foundation of drug discovery, even though few AI-designed drugs have reached the clinic. For pharma companies, embedding AI-driven workflows into R&D pipelines—using foundation models trained on internal genomics, transcriptomics, and proteomics data—may unlock previously intractable problems like neurological disease and accelerate candidate nomination timelines. However, commentators note the valuation cycle may have decoupled from clinical proof, meaning capital is chasing computational promise before real-world efficacy is proven.

   Foresite-backed Xaira Therapeutics, which launched in 2024 with more than $1 billion（約1600億円） in funding, is building virtual cell models to advance target discovery. Inductive Bio gained external validation in February by placing first in the OpenADMET-ExpansionRx blind challenge for predicting drug compound properties. The key differentiator for investors is whether AI can solve previously unsolvable problems and change the pace or probability of clinical success, not just model accuracy alone.

4. 4

   Anthropic launches drug discovery programs for neglected diseases

   Anthropic announced it is launching its own drug discovery programs targeting neglected diseases that traditional pharma and biotech firms consider unprofitable. The company will focus on early, preclinical-stage drug development. Anthropic also unveiled Claude Science, a new AI tool for research, and demonstrated early examples including a UCSF researcher using it to spot a viral contamination in minutes that his team had missed for an entire year. Pharma R&D timelines and success rates have long been constrained by information delays and operational bottlenecks. Novartis CEO Vas Narasimhan stated that AI tools could cut information and operational latency—which account for roughly 40 percent of total development time—potentially bringing drug development timelines down from twelve years to seven or eight years. Even modest improvements would matter: major pharma companies spend $150 to $200 billion（約32兆円） a year on R&D, and expanding the pool of treatable diseases could make previously unreachable drug targets viable.

   Claude Science analyzed 100 rare genetic diseases in under an hour and flagged 32 candidates for computational screening. Anthropic frames this drug discovery work as aligned with its nonprofit mission and as a way to build better AI models through firsthand experience in the sector. Other AI firms—including Deepmind (via Isomorphic Labs with Alphabet) and OpenAI—are also expanding into medicine and clinical tools.

5. 5

   Genesis AI model PEARL shows drug discovery can finally work—hitting real-world accuracy thresholds

   Genesis Molecular AI's PEARL model demonstrated on the OpenBind benchmark that it can accurately predict how small molecules bind to proteins, including modeling protein flexibility and induced-fit effects without fine-tuning on target-specific data. The model outperformed public competitors across evaluation metrics on 802 never-before-seen molecular complexes. Small-molecule drug discovery has long struggled because there are 10^60 drug-like molecules to search, and the properties that make a strong binder often conflict with those needed for the drug to reach its target in the body. PEARL's ability to model both ligand placement and protein adjustment together suggests that agentic drug-discovery loops—where AI iterates like a chemist, forming hypotheses and testing candidates—may now be practically feasible, potentially enabling 24/7 automated discovery cycles when paired with lab partners.

   The field has conventionally benchmarked poses at "2 Angstrom RMSD" accuracy, but Genesis argues that 1 Angstrom RMSD is the real threshold needed to correctly model molecular interactions like hydrogen bonds (which span only 0.6Å). PEARL's recent results suggest the community may be ready to abandon the weaker standard and pursue genuinely harder validation targets.

6. 6

   Meta's non-invasive brain-reading AI cuts word errors to 39%, closing gap with implants

   Meta researchers released Brain2Qwerty v2, which reconstructs typed sentences from brain signals measured outside the skull using magnetoencephalography (MEG). The system achieves a 39 percent word error rate, compared to 55 percent for previous methods, and requires ten times more training data than its predecessor to work without knowing exact keystroke timing. Invasive brain implants currently achieve below two percent word error rate, but they require surgery. This non-invasive approach—which works with portable room-temperature MEG sensors—offers a potential path toward clinical brain-to-text communication without surgery, though significant gaps remain and the system is not yet real-time capable.

   For the best participant, 28 percent of sentences decoded perfectly, and 47 percent contained at most one wrong word. The researchers found that collecting more recordings is a straightforward way to improve accuracy further, with no performance ceiling visible yet. Tests showed that even half the sensors deliver nearly full performance.