Alcon buys Neurotech for $550 million to gain eye cell therapy implant
Ophthalmic medicine maker Alcon is acquiring Neurotech Pharmaceuticals for $550 million, securing its commercial encapsulated cell therapy implant, Encelto.
News
FDA approves GSK hepatitis B drug Hibsago
The US Food and Drug Administration approved GSK's Hibsago (bepirovirsen), an antisense oligonucleotide for treating chronic hepatitis B. The therapy is designed to reduce viral load and hepatitis B surface antigens, giving patients an option to sustain viral suppression without remaining on indefinite treatment.
Reuters, also reported by FDAEndpoints News (subscription required)Fierce PharmaOpenAI acquires startup Orchestra Bio to expand life sciences agent suite
OpenAI has acquired Orchestra Bio, an early-stage developer of specialized software agents designed to coordinate clinical trial schedules, project monitoring, and drug development operations.
Yahoo Finance, also reported by The Information (subscription required)TechCrunchElectric Plant raises $11.2 million seed round to decode plant electrical signals with AI
Onto Ventures and Climactic co-led an $11.2 million seed round for Electric Plant, joined by Ponderosa, Endeavor Catalyst, Ludlow, Michigan Rise, The Fund, Schox VC, and Dug Song.
Lycia sets terms for $228 million IPO to fund allergy protein degraders
Lycia Therapeutics plans to raise $227.8 million in net proceeds, or roughly $228 million rounded in the report, by offering 14.7 million shares priced between $16 and $18 each.
FDA finalizes rules for apps paired with prescription drugs
The Food and Drug Administration published final guidance detailing how drugmakers must regulate software and mobile apps distributed alongside prescription medicines.
FDA, also reported by ReutersFierce PharmaCagriSema yields 25% average weight loss in women in Novo Nordisk analysis
Novo Nordisk reported that its experimental two-drug combination CagriSema delivered an average weight reduction of 25% among female participants in a post-completion study analysis.
Astellas commits up to $400 million in miniprotein alliance with AI Proteins
Astellas Pharma has partnered with Boston-based startup AI Proteins in a deal worth up to $400 million to develop tiny synthetic proteins designed by machine learning.
Endpoints News (subscription required)Rapport Therapeutics meets primary Phase 2 goal in bipolar mania
Rapport Therapeutics reported that RAP-219 met its primary endpoint in a Phase 2 trial treating bipolar mania.
Fierce Biotech, also reported by Endpoints News (subscription required)BioPharma DivePfizer halts Phase 2 trial of autoimmune antibody PF-06835375
Pfizer has terminated a Phase 2 trial testing its CXCR5-targeting antibody PF-06835375 in 61 patients with primary immune thrombocytopenia, an autoimmune condition causing low blood platelet counts.
Fierce Biotech, also reported by Endpoints News (subscription required)AstraZeneca teams with Oracle to use machine learning for trial recruitment
AstraZeneca has signed a non-binding agreement with Oracle to explore artificial intelligence applications designed to identify and recruit eligible clinical-trial patients earlier.
Fierce Biotech, also reported by Reuters
Preprints
Offline edge AI blocks pathogen synthesis by recognizing protein folds
Researchers built Capiti, an edge-computing screening system that monitors DNA synthesizer valve signals without an internet connection to abort production of toxic proteins.
Confidence scoring tells biologists when AI protein descriptions can be trusted
A sequence-to-text model called ProtTale writes functional summaries of proteins and estimates its own reliability using a calibrated confidence score termed pJay.
Modular software harness coordinates fifty protein-design tools into autonomous pipelines
RosettaCommons researchers released Chaperone, an open-source framework that integrates generative backbone models such as RFdiffusion3 with sequence optimizers like ProteinMPNN, structure predictors including AlphaFold3, and over 50 downstream design packages.
Deep-learning classifier sorts pneumococcal serotypes from raw bacterial genomes
Researchers built ALLCAPS, a neural network trained on 34,125 Streptococcus pneumoniae genomes that predicts bacterial serotypes directly from DNA sequences without comparing them against reference databases.
Computational framework reconstructs cell family trees directly from static RNA snapshots
Researchers built isCORAL, a computational framework that reconstructs clonal lineage histories and dynamic cell fates solely from static single-cell and spatial RNA sequencing data.
Convex optimization infers biochemical reaction rates without costly computer simulations
Researchers introduced simulation-free approximate Bayesian computation (SFABC), a statistical framework that infers reaction parameters in stochastic biochemical networks.
Tiny language model trained only on plant peptides matches general protein models forty times its size
A 170,000-parameter masked language model trained exclusively on 1,469 plant signalling peptide precursors matched the peptide family classification accuracy of an 8-million-parameter ESM-2 model.
Simple additive measurements surpass protein foundation models after 100 lab tests
A benchmark evaluating 95 computational protein predictors across 217 assays and 2.47 million variants from ProteinGym examined whether AI models actually improve experimental design decisions under realistic lab budgets.
Unified sequence model maps gene activity across single cell states and mammal species
A deep-learning model called Bolero conditions its sequence predictions on cell-state embeddings from large single-cell atlases to forecast chromatin accessibility and RNA abundance directly from DNA.
Iterative wet-lab feedback drives AI agents to design working DNA aptamers
An automated AI agent pipeline iteratively designed target-binding aptamers, which are short synthetic nucleic acid strands, using experimental data from successive 1-to-2-week laboratory testing rounds.
Continuous thought architecture models brain tumor cell trajectories and gene switches
Researchers built scCTM, a machine-learning framework adapting Sakana AI's Continuous Thought Machine to reconstruct developmental trajectories and state transitions in aggressive glioma cells.