AI Heals

OpenAI unveiled GPT-Rosalind, the first model in a new family aimed specifically at biology, drug discovery and translational medicine. The system is trained to read scientific papers, query laboratory databases, design experiments and generate biological hypotheses, and ships with a Life Sciences research plug-in for Codex that connects to more than 50 scientific tools and data sources. OpenAI reports stronger results than its general-purpose flagship on biochemistry, experiment design and tool-use benchmarks, and says the model out-scored 95% of human scientists on a blind RNA prediction test set built by gene-therapy laboratory Dyno Therapeutics. Made available in research preview to qualifying enterprise users, with Amgen, Moderna and the Allen Institute named among early adopters; signals a shift towards purpose-built domain models alongside frontier general-purpose systems.OpenAI announcement

Roche announced an expansion of its hybrid-cloud AI infrastructure, adding 2,176 NVIDIA Blackwell GPUs to bring its total fleet above 3,500. The on-premises systems span the United States and Europe, and feed into Roche's Lab-in-the-Loop pipeline through NVIDIA's BioNeMo platform, where experiments and AI models iterate together across discovery, development and manufacturing. Marks the largest announced AI compute deployment by a pharmaceutical company.Roche Media Release

Engineers at USC Viterbi unveiled a fully automated AI tool that detects cancer cells in a blood sample in as little as ten minutes, with no human in the loop. The system scans the sample cell by cell and isolates rare circulating tumour cells that human pathologists routinely miss, returning a result in time for the same clinic visit rather than after days of laboratory work. Where most liquid-biopsy AI to date is a research demo, the USC tool is built as a deployable diagnostic — fast enough for the clinic, accurate enough for triage, and automated enough to run without specialist staff. If it survives validation, the time-to-result curve for cancer screening collapses from days to a single appointment.USC Viterbi announcement

Johns Hopkins researchers reported MIGHT, an AI method for detecting advanced cancer from a blood sample using aneuploidy-based features. Tested on 1,000 patients (352 with advanced cancers, 648 without), MIGHT delivered 72% sensitivity at 98% specificity — a markedly better reliability profile than previously published AI blood-test cancer detectors. The performance is significant because the failure mode that has dogged this field is false positives: a screening tool that cries cancer too often forces unnecessary biopsies and erodes trust. MIGHT's specificity bench means a positive call now carries enough weight to actually drive next steps, moving liquid biopsy AI closer to routine clinical use.Johns Hopkins announcement

MIT researchers used generative AI to design entirely new chemical compounds effective against drug-resistant bacteria. Unlike previous AI-assisted antibiotic discovery that screened existing libraries, this approach generated novel molecular structures from scratch, targeting specific bacterial vulnerabilities identified by the model. Demonstrated that AI can move beyond screening known molecules to designing entirely new ones, opening a generative approach to antibiotic development against superbugs.MIT News

The FDA granted de novo authorisation to Clairity Breast, the first AI tool cleared to predict a woman's five-year risk of developing breast cancer using only a standard screening mammogram. The model reads subtle imaging patterns invisible to radiologists and outputs a personal risk score that can be used to escalate screening, push for MRI, or reassure low-risk women. It is a regulatory landmark — the first authorisation in the United States for predictive (rather than diagnostic) AI in oncology — and shifts AI's clinical role from detecting cancer that already exists in an image to predicting whether cancer will appear in a given patient. The decision opens a route for similar long-horizon risk models in cardiology, dermatology and ophthalmology.BCRF breakthroughs report

Sydney tech entrepreneur Paul Conyngham used ChatGPT to identify candidate tumour antigens and AlphaFold to model the resulting proteins, producing the design for a personalised mRNA cancer vaccine for his dog Rosie in under two months. UNSW researcher Pall Thordarson synthesised and administered the vaccine, and by February 2025 Rosie's leg tumour had shrunk by 75%. The first known case of an AI-designed personalised cancer vaccine being successfully used in a live animal, demonstrating that the same mRNA platform behind COVID-19 vaccines could be rapidly adapted to individual cancer profiles at low cost.Fortune

The 2024 Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for foundational work on artificial neural networks. The Chemistry prize went to David Baker for computational protein design, and to Demis Hassabis and John Jumper of DeepMind for AlphaFold — an AI system that solved the 50-year protein structure prediction challenge. It was the first time AI methods earned Nobel recognition in two disciplines simultaneously. Established AI as a Nobel-calibre scientific methodology, recognising both the foundational neural network research and the applied scientific breakthroughs it enabled.Nature Machine Intelligence

DeepMind released AlphaFold 3, expanding beyond protein structure prediction to accurately model interactions between proteins and DNA, RNA, and small-molecule drugs. The system achieved a 50% accuracy improvement over AlphaFold 2, enabling prediction of the molecular interactions that underpin most biological processes. Extended AI's reach from predicting single protein shapes to modelling the full complexity of molecular biology, accelerating drug discovery and fundamental research across the life sciences.Nature

MIT researchers used deep learning to identify an entirely new structural class of antibiotic compounds effective against methicillin-resistant Staphylococcus aureus (MRSA). The AI model not only found the candidates but also explained its reasoning, revealing the chemical substructures responsible for antimicrobial activity — a key advance in interpretable AI for drug discovery. Showed that AI-driven drug discovery can be both effective and interpretable, addressing a major criticism of black-box approaches in pharmaceutical research.MIT News

DeepMind released predicted structures for nearly all 200 million proteins known to science — the entire protein universe catalogued in UniProt. The AlphaFold Protein Structure Database expanded from 1 million to over 200 million entries in a single release, freely accessible to researchers worldwide. Gave every biologist on Earth instant access to protein structures that would have taken centuries to determine experimentally, fundamentally altering the pace of biological and medical research.Nature