All articles tagged with #drug discovery
Researchers have developed a human-induced pluripotent stem cell (iPSC) spheroid model that accurately simulates myelin damage and repair in the central nervous system. This system allows for the study of remyelination processes, including microglial response and oligodendrocyte regeneration, offering a new platform for drug screening in multiple sclerosis.
AI systems are evolving from chatbots into autonomous agents that generate hypotheses and design experiments. While these tools speed up research, experts emphasize that human judgment remains critical for framing questions and verifying results.
Researchers at Caltech and the University of Pittsburgh have developed a new biocatalytic method to synthesize chiral oxazolidinones, key scaffolds in next-generation antibiotics, directly from unactivated alkenes using engineered haemproteins.
London-based AI startup Basecamp Research has secured a $140 million Series C funding round led by S32, with participation from Nvidia, Anthropic’s Anthology Fund, and the NATO Innovation Fund. The capital will support the expansion of its EDEN AI models, which analyze genetic data from diverse global ecosystems to design antibiotics and gene therapies. The company aims to move its six therapeutic candidates from lead optimization toward clinical trials, focusing on in vivo cell therapies and antimicrobial peptides.
A Stanford-led study built a ‘Virtual Biotech’ of up to 37,000 AI agents that analyzed results from more than 55,000 clinical trials and gene-activity data. The system identified CD276 as a promising lung-cancer target and proposed an antibody-drug conjugate strategy, demonstrating how AI could accelerate drug discovery, though the approach has not been experimentally or clinically validated and requires human oversight.
An AISB Network effort trained OpenFold3 on about 20,000 proprietary protein–ligand structures from pharma companies, producing a model that surpasses AlphaFold‑like systems trained only on public data and those trained on single‑company data; the proprietary data remained private and the study has not yet been peer‑reviewed.
Researchers generated over 38 million time-resolved proteomic measurements from systematically perturbed breast cancer cell lines and built ProteinTalks, a scalable virtual cell model that learns transferable dynamic representations from proteome trajectories to predict drug efficacy and synergy, discover new drug combinations, probe proteins linked to drug resistance, and stratify patient responses. The model shows robust transferability to patient-derived organoids and clinical biopsies, generally outperforming benchmark methods, and data/code are publicly available to enable in silico drug discovery.
A solo biochemist claims to have designed via ChatGPT and synthesized PAC-3310, a schizophrenia drug, in a garage, modeling it after a recently approved antipsychotic. Experts emphasize that drug development requires extensive preclinical work and a costly, multi-year FDA process, making DIY creation and testing highly unlikely to reach approval and highlighting risks of overreliance on AI for complex therapeutics.
An AI-designed drug from Insilico Medicine, rentosertib, reduced biological aging markers across six aging clocks in a 12‑week study with 42 participants, suggesting AI may modulate aging biology. However, aging clocks are debated, and this is an early-stage finding with many trials and regulatory hurdles still to come.
Insilico Medicine used generative artificial intelligence to design rentosertib, a small-molecule drug for a chronic lung condition. New clinical-trial data reported in Nature Biotechnology indicate the AI-designed drug reduced biological aging markers across six aging clocks, suggesting a potential to slow aging—but further research is needed to confirm safety, efficacy, and applicability beyond the lung disease context.
AI giants like Anthropic are pushing deeper into health care and biology to repair their public image, diversify revenue, and fund the expensive compute behind model training, with collaborations such as Nvidia/Lilly and Isomorphic Labs; while early results may take years, the move faces public skepticism, regulatory hurdles, and concerns about costs and safety, and its success will depend on delivering tangible patient benefits.
Scientists published the largest map to date of how autism-linked genetic mutations rewire brain protein networks, using AI and lab-grown brain organoids to identify over 1,800 protein–protein interactions (87% new) and prioritize mutations with AlphaFold. The resulting molecular wiring diagram could shift drug development from targeting individual genes to stabilizing disrupted protein complexes, with three therapeutic programs already underway, though translating findings into treatments will take time and may not apply to all autism forms.
UCSF researchers built a molecular map linking 100 autism-risk genes to about 1,800 protein interactions, most previously unreported, and used AlphaFold to pinpoint direct connections. They found a FOXP1-FOXP4 interaction that drives autism-like traits in human organoids, suggesting a new drug-discovery blueprint—targeting protein networks rather than genes alone—to identify potential therapies, with potential pathways for industry collaboration.
Scientists have produced the largest map of autism-related protein interactions by using AI and lab-grown brain organoids to show how genetic mutations rewire brain networks, creating a blueprint of protein wiring rather than a gene-by-gene target; the findings point toward therapies that stabilize disrupted protein complexes, with three drug-development programs already underway, though translating these insights into treatments will take time and may not cover all autism forms.
Generative AI can speed up biology research and drug screening, but it can also generate plausible yet false biological signals. Experts say AI outputs should be treated as testable ideas or synthetic data rather than definitive findings, and must be independently verified in real experiments. The risk varies by use, with cases like AlphaFold 3 illustrating how AI errors can distort evidence if not carefully checked.