The paper presents a data-efficient AI workflow, Bayesian optimization implemented in the AutODEx framework, for accelerating the discovery of thermostable mRNA-lipid nanoparticle vaccines. Candidate formulations were optimized toward stability, using a SARS-CoV-2 RBD sequence deposited at GenBank (OP839194). Datasets and source code are publicly released on Zenodo and GitHub.
FRANK, a neural network classifier for pan-cancer pediatric tumors, distinguishes 181 childhood tumor and normal tissue types from RNA-seq data. Trained on 11,467 curated transcriptomes with data augmentation and a hierarchical penalty, it detected tumor types from as few as three training examples and reached 98.3% overall accuracy across nine validation datasets totaling 16,398 samples. It outperformed other published pediatric pan-cancer RNA-seq classifiers on an external dataset and remained accurate on long-read, noisy and low-quality data.
Why it matters: It sets a new accuracy benchmark for RNA-seq-based classification across 181 pediatric tumor types, including with sparse and low-quality data.
SHERLOCK is an interpretable deep generative framework that represents genetic, pharmacological, and experimental perturbations as structured interventions on a latent baseline cellular state within a structural causal model. It learns correlated and sparse perturbation representations, organizes perturbations by shared transcriptional responses, and supports counterfactual estimation of downstream transcriptional effects, condition-dependent response analysis, and compositional modeling of combinatorial perturbations. Across genome-scale CRISPR, chemical, and spatial perturbation datasets, it recovers perturbation relationships concordant with known pathways and pharmacological properties and predicts held-out perturbation combinations.
Genentech will pay Earendil Labs $55 million upfront to discover and develop several bispecific antibody cancer therapies, in a deal that could exceed $1.5 billion including milestones, plus tiered royalties. Earendil leads antibody discovery and early clinical research on pre-agreed target combinations, while Genentech handles global development and commercialization; oncology indications have not been named. Roche also struck a $70 million upfront deal with Atavistik Bio the same day.
Enveda, a company applying artificial intelligence to drug discovery, has raised $311 million from AI and biotech investors. The financing, announced via Business Wire, is intended to support its effort to modernise pharmaceutical research and development. No trial data, pipeline specifics or timelines were disclosed alongside the raise.
An analysis of 1.3 million chemical reactions from patents across four decades shows hazardous solvent use rose from about 60% of reactions in 1976 to above 70% by 2016. Helen Sneddon of the University of York and Maarten Dobbelaere of the Swiss Federal Institute of Technology in Lausanne used the AI system Rxn-INSIGHT on a US Patent and Trademark Office database, publishing the survey this month in Angewandte Chemie. When regulation restricts a toxic solvent, chemists tend to switch to another hazardous liquid rather than a safer alternative, and use of trifluoroacetic acid, a PFAS, has risen substantially; the database ends in 2016 and does not measure solvent quantities.
An FDA expert committee voted on September 23 that the clinical benefits of Galleri, a multi-cancer early-detection blood test, outweigh its risks. The vote marks a key step toward formal FDA review of the product, which is positioned as a blood draw for early cancer screening. The agency's decision process now moves forward following the panel's endorsement.
CMAC, a deep learning framework built from 18 murine single-cell RNA-seq datasets, assigns absolute transcriptional maturation age to cardiomyocytes across platforms and experimental conditions. An XGBoost regressor on its batch-invariant latent space predicted chronological age with leave-one-dataset-out MAE of 4.09 days (Pearson r = 0.903), and in an independent P14 cohort of 98,163 cells achieved mouse-level MAE of 0.33 days. Applied to PGC1/beta double-knockout mice, it quantified maturation deficits of 4.77 to 8.60 days across P7 to P28.
A software framework called the Model Hardware Standard (MHS), developed by Anthropic and the Howard Hughes Medical Institute's Janelia Research Campus, connects laboratory instruments from different vendors into a single system that AI agents can control to run experiments. A team testing it at Carnegie Mellon University set up a functioning experiment in hours, a process that can easily take months. Unlike the SiLA consortium's existing standard, MHS links instruments directly to AI agents.
unZipro is a geometric deep learning framework that adapts general protein design models to specific protein families using meta-learning to navigate family-specific fitness landscapes. It enabled rapid in silico protein evolution across gene editing, luciferase reporting, transcriptional regulation, and antiviral resistance with minimal experimental screening. The work simplifies protein evolution by cutting the screening burden required to tailor designs to a chosen family.
The AlphaFold Protein Structure Database has added more than 8,000 virus protein dimers from 23 virus families that infect humans, including monkeypox, measles and hepatitis B viruses. Researchers analysed sequences of 41,774 proteins from around 2,800 viruses and used AlphaFold2 to predict roughly 40,746 homodimers and nearly 1.7 million heterodimers, of which only 2,749 homodimers and 5,279 heterodimers met accuracy standards for inclusion. The update, launched on 24 September with a pandemic preparedness portal, fills a viral blind spot in a resource with more than three million users.
The U.S. Department of War has awarded Flagship Pioneering a $47 million contract over five years to develop medical countermeasures through its OnTarget program. Initial work goes to two Flagship-founded biotechs: Apriori Bio, whose AI-driven Octavia platform predicts viral evolution for vaccine design, and Prologue Medicines, whose Delve platform identifies immunomodulatory proteins. Additional Flagship projects may follow.
Boehringer Ingelheim has expanded its collaboration with Seattle-based genomics biotech Variant Bio to cover cardiovascular disease, on top of the original January deal worth $120 million for cardiorenal and kidney disease. Variant will receive an additional upfront payment and is eligible for license and milestone payments totaling more than $250 million. The partners will use Variant's AI-powered Inference platform to identify genetically supported targets for potential first-in-class cardiovascular treatments.
A graph neural network applied to single-cell data from 584 brain donors with Alzheimer's disease and controls identifies differential patterns between disease and control. The model links specific patterns to cognitive resilience and to the insurgence of depression in patients with AD, giving a data-driven way to parse disease phenotypes at single-cell resolution.
Nature Medicine published on 23 September 2026 a piece on scaling a deep learning clinical tool from a single hospital to more than a million screened patients across India, Thailand and Australia. The three distinct settings yielded cross-cutting lessons on deployment. The authors suggest these insights can inform wider global expansion of healthcare artificial intelligence.
A study shows that an AI-designed IL-21 mimetic can convert transient responses to KRAS inhibitors into durable remissions in pancreatic cancer through CD4 T cells, independent of CD8 T cells and tumor MHC class I expression.
Why it matters: The study reveals how an AI-designed IL-21 mimetic turns CD4 T cells into mediators of KRAS inhibitor efficacy in pancreatic cancer, offering a new angle on overcoming resistance.
Iambic Therapeutics, a San Diego AI drug discovery biotech, has filed with the SEC to go public after raising a $100 million series B in 2023 and a similar round in November 2025. IPO proceeds would chiefly fund IAM1363, an oral brain-penetrant HER2 inhibitor in a phase 1 study in advanced HER2-altered solid tumors, with a registrational trial planned for early 2027. The company had $207.9 million in the bank at the end of June and partnerships including a Takeda deal worth up to $1.7 billion.
Esophageal cancer detection through noncontrast computed tomography is enhanced by a new artificial intelligence tool called Esophageal AI-Guided malignant Lesion Evaluation. In a large-scale study, the tool achieved high sensitivity and specificity across diverse settings.
Evo2, a genomic foundation model, was used to embed short variants, fusions and rearrangements from 102 patients with soft tissue sarcoma and 129 TCGA-SARC cases into patient-level genomic representations for Cox survival modelling. Adding Evo2 to SARCULATOR raised the OS C-index from 0.620 to 0.770 in the study cohort and from 0.595 to 0.716 in TCGA-SARC; in TCGA-SARC it also improved CINSARC-based OS prediction from 0.476 to 0.653. These descriptive point estimates support altered tumor sequence context as an independent prognostic layer for sarcoma risk stratification.
Ataraxis Breast CTX, a causal multi-modal AI model combining clinical variables with H&E histopathology, stratified residual recurrence risk in 556 node-positive HR+/HER2- breast cancer patients from the UNIRAD trial. Five-year disease-free survival was 93% in the low-risk group versus 80% in the high-risk group, and CTX-benefit predicted everolimus benefit in an exploratory multivariable analysis (interaction p = 0.01). The model may inform selection of patients for adjuvant treatment escalation.
