The front page of AI-driven biology
Wednesday, October 21, 2026Wed, Oct 21
Updated 10:02 PM PT
Rezera, formerly known as NodThera, has filed to list publicly under the ticker RZRA to fund clinical development of its lead oral NLRP3 inhibitor, ruvonoflast.
The agency cleared the drug for adults with immunoglobulin A nephropathy, an autoimmune disorder that attacks the filtering units of the kidneys and can progress to organ failure.
The Swedish private equity firm submitted a binding takeover proposal offering more than $50 per share for the life-sciences consumables and molecular diagnostics supplier, Bloomberg reported.
A genome-wide association study led by researchers at Johns Hopkins University mapped 127 independent genetic regions tied to clinical obesity, 63 of which showed no significant connection to standard body mass index (BMI).
Draft outbound investment rules from the U.S. Department of the Treasury are expected to tighten scrutiny on biotechnology transactions in China, according to reporting by Endpoints News.
The all-cash transaction transfers the Foresight carrier screen, Prequel non-invasive prenatal test, SneakPeek, and FirstGene assays to expand Labcorp's reproductive diagnostics footprint against rivals Natera and Quest Diagnostics.
Hologic has agreed to acquire Barcelona-based precision oncology developer Reveal Genomics to broaden its breast cancer diagnostics portfolio.
Computational biologists paired an artificial neural network with simplified protein simulations to model how intrinsically disordered proteins fold transiently.
Researchers introduced Pinal, a 16-billion-parameter artificial intelligence model that generates functional proteins directly from natural-language descriptions.
Benchmarking five histopathology foundation models on 3,528 whole-slide images from Barrett’s esophagus patients showed that combining them improved detection accuracy for esophageal adenocarcinoma (AUROC 0.909) and morphological subtyping (AUROC 0.907).
A retrospective study of two FDA-cleared Aidoc algorithms across 67,869 CT scans found that while the tools maintained high specificity (99.1% for pulmonary embolism and 99.8% for incidental cases), their sensitivity dropped significantly compared to clearance benchmarks.
Researchers developed DynaVelo, a generative neural ordinary differential equation model that maps simultaneous gene expression and transcription factor motif dynamics from single-cell data.
A system named PEARL connected general-purpose large language models to a liquid-handling robot and a camera, allowing the software to independently plan, execute, and interpret wet-lab chemistry tasks.
Carbon-Annotator, a 1.2-billion-parameter neural network, predicts protein-coding genes directly from raw DNA sequences across animals, plants, fungi, and single-celled eukaryotes without needing external reference templates.
Researchers developed CPMsi, a computational framework that aligns raw spatial mass spectrometry imaging data, detects similar molecular zones, and compresses large tissue scans into representative spectra.
Researchers generated synthetic protein binders from scratch targeting three phospho-tau sites and pyroglutamate-modified amyloid-beta (N3pG-Abeta) with sub-nanomolar affinity.
Semi-supervised learning using enzyme-substrate records improves predictive accuracy on benchmark datasets clustered at 40% sequence identity.
A multimodal transformer incorporating mixture density networks predicts the developmental migration paths of mouse astrocytes more accurately by incorporating vascular environment maps.
BiNVaST couples a dual-branch variational autoencoder with synthetic data generation using singular value decomposition and optimal transport to improve cell classification in subcellular spatial data.
Researchers released an open dataset profiling the proteins, genes, and tissue images of 1,000 untreated colorectal cancer cases, creating the largest proteomic cohort for this disease to date.
Researchers trained hEVO, a specialized genomic foundation model, on 4.784 billion haplotype sequences reconstructed from whole-genome data across 63,809 individuals to capture natural human genetic variation.
Magnolia coordinates computational protocols from docking to quantum chemistry using a three-tier memory structure, saving session notes into project-level entries and human-approved global rules.
Researchers introduced Tessera, a representation learning framework that encodes local tissue microenvironments using relative cellular orientations rather than simple distance or rigid tissue coordinates.