Skip to content

#AI制药

Today 0 items
Oct 15Thu
  1. biospace.com69

    AstraZeneca bulks up MASH, cardio work in up to $1.5B AI partnership with insitro

    AstraZeneca will pay insitro up to $1.5 billion to discover drug targets for metabolic dysfunction-associated steatohepatitis and cardiomyopathies using insitro's Virtual Human AI platform. Insitro receives $40 million upfront and up to $1.46 billion in milestones plus royalties, while AstraZeneca advances any candidates through development. The deal adds to AstraZeneca's MASH work, which includes the Phase 1 anti-fibrotic asset AZD2389.

    Why it matters: A large pharma-AI target discovery deal in MASH and cardiomyopathies, indications where AstraZeneca has recent setbacks.

  2. Nature76

    Artificial intelligence-enabled clinical phenotyping of human brown fat.

    Brown adipose tissue (BAT) presence is linked to lower prevalence of cardiovascular and psychiatric diseases across a cohort of 101,243 patients from a large cancer centre. Machine learning quantification of BAT activity in roughly 8,700 patients showed higher activity, especially in the axillary depot, correlates with lower odds of cardiometabolic diseases and healthier white fat distribution. Angiotensin receptor blocker use associated with BAT presence, and losartan increased brown adipocyte activity and whole-body energy expenditure, supporting drug repurposing to activate BAT in humans.

    Why it matters: AI phenotyping of 101,243 patients links brown fat to reduced cardiometabolic disease and identifies losartan as a candidate BAT activator.

  3. Nature77

    A microexon regulatory code reveals autism-linked genetic variation.

    An expanded map of neuronal microexons, built from extensive transcriptome profiling, adds hundreds of new microexons, including subtype-specific ones and others that trigger nonsense-mediated mRNA decay. A machine-learning sequence code predicts neuronally spliced microexons and identifies regulatory elements around them. Applied to autism, the code pinpoints rare splicing-altering variants, mostly in idiopathic individuals, that associate with autism status and converge on synaptic and other autism-linked genes.

    Why it matters: It links autism-associated genetic variation to a splicing code for neuronal microexons.

  4. gov.uk51

    MHRA 2030 Strategy to modernise patient-centred regulation for the era of AI and personalised medicines

    The Medicines and Healthcare products Regulatory Agency (MHRA) has set out plans to modernise patient-centred regulation in preparation for an era of AI and personalised medicines. The strategy, running to 2030, aims to translate scientific and technological innovation into better health outcomes for patients across all four nations of the UK. Implementation of the strategy's measures is the next step for the agency.

    Why it matters: MHRA's 2030 strategy will shape how AI and personalised medicines are regulated in the UK.

  5. Nat Biotechnol63

    Multimodal learning enables chat-based exploration of single-cell data.

    CellWhisperer links transcriptomes to text through a multimodal embedding trained by contrastive learning on 1 million RNA sequencing profiles with AI-curated descriptions, feeding a large language model that answers questions about cells and genes in natural-language chats. Benchmarked for zero-shot prediction of cell types and biological annotations, it supported discovery in a meta-analysis of human embryonic development and is integrated with the CELLxGENE browser for combined graphical and chat exploration of gene expression.

  6. biopharmadive.com61

    Iambic, an AI drug discovery specialist, banks $150M in an IPO

    Iambic Therapeutics, an AI drug discovery startup, raised $150 million in an IPO, selling close to 9.4 million shares at $16 each and trading on the Nasdaq from Thursday under the ticker "IAM." The company has drawn $462 million in private funding since 2019 and will use the proceeds to advance its first drug, IAM1363, an oral tyrosine kinase inhibitor for HER2-positive solid tumors that showed anti-tumor activity and encouraging safety in a Phase 1 study, with a registrational trial possibly beginning in 2027. Behind it are IAM217, targeting KIF18A in ovarian and triple-negative breast cancers, and a CDK2/CDK4-targeting prospect, both potentially entering testing next year.

    Why it matters: The offering adds to a wave of AI drug discovery startups testing whether public investors will back the technology.

  7. medrxiv.org40

    Choroid plexus asymmetry on routine MRI as an imaging signature of breast cancer brain metastasis

    Choroid plexus asymmetry on routine contrast-enhanced brain MRI is associated with breast cancer brain metastasis. In 1,000 women with breast cancer, 190 with brain metastasis, a deep-learning-derived Choroid Plexus Asymmetry Index was elevated in metastasis cases (median 0.088 vs 0.059; adjusted OR 1.58 per +1 SD, 95% CI 1.32-1.88), independent of peripheral inflammation markers, histological grade, tumour laterality and metastasis side. Total choroid plexus volume did not differ between groups, pointing to asymmetry rather than size as the relevant signature.

  8. nature.com71

    Artificial intelligence-enabled clinical phenotyping of human brown fat

    AI-enabled quantification of brown adipose tissue (BAT) activity in humans links higher BAT activity to a healthier metabolic profile and lower rates of psychiatric disease. The same approach identified existing pharmacologic agents that could be repurposed to activate BAT in humans. The work provides a scalable way to phenotype BAT and connect it to clinical outcomes and drug targets.

    Why it matters: It shows AI-based BAT phenotyping at scale, linking brown fat to metabolic and psychiatric outcomes and to repurposable drugs.

  9. nature.com80

    Computational design of metalloproteases

    RFdiffusion was used to design proteins de novo with water-activating zinc-based active sites positioned next to a peptide substrate binding pocket. The resulting proteases were highly active and programmable, with the authors pointing to potential applications in therapeutics, biotechnology and bioremediation.

    Why it matters: De novo design of active, programmable metalloproteases extends computational protein design to catalysis.

  10. nature.com82

    De novo design of cysteine proteases

    Designed cysteine proteases from scratch using a computational design pipeline built on RFdiffusion2 active-site scaffolding, ProteinMPNN sequence design, and AlphaFold 3 prediction. Candidate enzymes were characterized by crystallography, biochemical kinetic assays, and mammalian-cell imaging, extending the de novo enzyme design approach to the cysteine protease class.

    Why it matters: De novo design reaches cysteine proteases, a class previously outside computationally designed enzymes.

Oct 14Wed
  1. biospace.com63

    Zealand pens up to $1.2B deal with Generate for AI-designed obesity, metabolic drugs

    Zealand Pharma will pay Generate:Biomedicines $30 million upfront, up to $35 million in early preclinical milestones and nearly $1.2 billion in development, regulatory and commercial milestones, plus tiered royalties, to access Generate's generative AI platform for designing protein therapeutics for obesity and metabolic diseases. Generate handles research through candidate selection; Zealand leads clinical development, regulatory and commercialization. The deal follows mixed Zealand readouts for petrelintide and survodutide.

    Why it matters: Large AI drug-discovery alliance in obesity adds Generate's platform to Zealand's metabolic pipeline after mixed trial results.

  2. Nat Med71

    MAGIC: an international network for evaluating generative artificial intelligence in global health.

    The authors establish MAGIC, an international network for evaluating generative artificial intelligence in global health. The network brings together collaborators across institutions in Asia, Europe, Africa and the Americas, including groups at Duke-NUS Medical School, Harvard, Tsinghua and the World Health Organization Regional Office for South-East Asia, with a focus on the ethics, regulation and standardization of AI for health.

    Why it matters: Sets out an international framework for evaluating generative AI in global health, relevant to clinicians and researchers deploying these tools.

  3. biorxiv.org65

    Decoding super-enhancer regulatory grammar and pathogenic gain-of-function variants directly from DNA sequence

    Borzoi-SE is a deep learning framework that identifies super-enhancers directly from DNA sequence across 90 human lineages, without epigenomic profiling. It decoded the functional hierarchy of the β-globin super-enhancer, prioritizing HS2 as the dominant constituent, verified by CRISPR interference. Trained without variant-effect labels, it identified gain-of-super-enhancer variants, including a single-nucleotide variant at the CHRM2 locus that created a de novo glucocorticoid receptor motif and nucleated a cardiac super-enhancer linked to resting heart rate and PR-interval variation.

  4. medrxiv.org41

    MRI Biomarkers for Brain Metastases in Lung Cancer Using Machine Learning

    MRI-derived radiomic features improved prognostic stratification beyond clinical variables in 103 patients with brain metastases from lung cancer. The combined clinical-imaging DeepSurv model reached a concordance index of 0.728 for overall survival, and risk groups separated median overall survival at 26.7 versus 9.1 months. Imaging-only high-risk patients were less likely to have tumor-infiltrating lymphocytes and PD-L1 expression of at least 50%, linking the biomarkers to the tumor immune microenvironment.

Oct 13Tue
  1. biospace.com56

    Pfizer pens $500M+ deal with Enodia to destroy disease-causing proteins

    Pfizer has signed a deal with Paris-based Enodia Therapeutics worth up to $500 million, covering an upfront payment plus research, development and sales milestones, with tiered royalties on net sales. Enodia will screen, optimize and develop small-molecule inhibitors of the Sec61 membrane protein complex, while Pfizer handles clinical studies, manufacturing, regulatory and commercial work. Target indications are not yet specified. Pfizer's venture arm had co-led Enodia's €20.7 million ($25 million) seed round in January.

    Why it matters: Pfizer commits over $500 million to a Sec61-targeting platform less than a year after the biotech's launch.

  2. biorxiv.org59

    A high-resolution protein-RNA binding atlas for structure-aware interaction prediction and post-transcriptional regulatory modeling

    A high-throughput CLIP-seq platform, eclair CLIP, maps binding sites for 67 endogenously tagged RNA-binding proteins in HEK293T cells, yielding a transcriptome-scale atlas of post-transcriptional regulons linked to RNA half-life. Adding ecCLIP data to the Borzuki sequence-to-function model improved RNA half-life prediction and assessment of 3' UTR variant effects on RNA stability. A structure-aware neural network, Pythia, predicted RBP binding from sequence and structural context, with mutation libraries in reporter CLIP assays confirming predicted structure-dependent binding for five RBPs and allele-specific variant effects.

    Why it matters: It provides a 67-protein RBP binding atlas plus models that improve prediction of RNA stability and 3' UTR variant effects.

Oct 12Mon
Oct 11Sun
  1. doi.org58

    Catching enhanced flexibility in antibody-antigen recognition: critical evaluation of physics-based and AI-driven methods

    The affinity-matured Hy104X antibody gains high affinity by actively destabilizing its CDRH1 loop, capturing mobile lysozyme variants through entropy-buffered conformational selection. Both MM/PBSA and AI-driven structure predictors failed to rank affinities for more flexible systems, but adding structural templates to Chai-1 improved correlation with experimental affinities. The findings point toward coupling AI screening with physics-based evaluation for antibody engineering.

  2. biorxiv.org62

    Spatiotemporal AI modeling of gene regulation and variant effects in the developing human brain

    Seal, an interpretable AI transfer learning framework, models gene expression and variant effects with spatiotemporal resolution across 26 brain regions, 30 cell types, and seven developmental stages of the human brain. Applied to GWAS, it identifies cell types and developmental windows relevant to six neuropsychiatric conditions, and in Simons Simplex Collection autism whole genome data it uncovers a significant burden of de novo regulatory variants in transient fetal excitatory neurons.

    Why it matters: It links disease-associated variants to specific cell types and developmental windows in the developing brain.

  3. biorxiv.org70

    A unified framework for quantitative splicing and transcript prediction

    SpliceAI2 is a deep learning model trained end-to-end on splice site, splice junction, and transcript usage data across multiple mammalian species, and it reconstructs complete transcripts directly from genomic sequence. It improves detection and quantification of splice-altering variants across benchmarks covering cryptic splicing, splicing QTLs, and massively parallel reporter assays. In a rare disease cohort, predicted splice-altering variants account for 15% of the excess genetic burden, with about half in deep intronic regions likely missed by exome sequencing.

    Why it matters: SpliceAI2 quantifies splice-altering variant effects genome-wide, including deep intronic variants exome sequencing misses.

Oct 10Sat
  1. Lancet72

    Conversational diagnostic artificial intelligence in ambulatory primary care: a prospective feasibility study.

    A patient-facing conversational AI, AMIE, was used by 114 patients up to 5 days before urgent primary care appointments, with 98 completing both the AI interaction and the physician visit. Zero prespecified safety stops occurred, and clinical evaluators rated conversations favourably in 87-100% of cases across 17 criteria. PCPs who reviewed the transcript before the visit found it helpful in 33 of 44 cases and said it might have changed their behaviour in 25 of 44.

    Why it matters: First real-world feasibility data on a patient-facing conversational AI in urgent primary care, with zero safety stops.

Oct 9Fri
  1. medrxiv.org50

    A spatial AI-based biomarker with independent prognostic value in early low- and intermediate- risk breast cancer

    Clusters of high-grade morphology at the invasive tumour front, detected with the ViT_DG vision-transformer on whole slide images, independently predicted outcome in ER-positive/HER2-negative/lymph-node-negative Grade 1-2 breast cancer. In a Swedish multicentre cohort (6106 patients; 1403 test patients in the primary subgroup), cluster positivity carried a 2.30-fold increased recurrence risk, and cluster-negative patients reached a 10-year breast-cancer recurrence-free interval of 95.3%. The marker also flagged risk (HR 2.86) among patients scored low-risk by the global AI biomarker.

  2. biorxiv.org56

    Cerberus: bidirectional state space blocks improve accuracy and efficiency of regulatory sequence models

    Replacing Borzoi's transformer blocks with Hydra, a bidirectional Mamba-2-based state space block, yields regulatory sequence models that run 5.7-fold faster on 8,192-token inputs and predict binned coverage and fine-mapped eQTLs more accurately. Scaled into Cerberus, an ensemble of eight models over 786 kb of input, the architecture beats Borzoi on GTEx v11 eQTL, sQTL and paQTL benchmarks, with mean eQTL AUPRC of 0.692 versus 0.668 across 48 tissues and effect-size Spearman ρ of 0.379 versus 0.321. Trained blocks show interaction range growing with depth, from a 0.9 kb median half-life in the first block to 79 kb in the seventh.

  3. biorxiv.org50

    Fine-tuning genomic models on rare splicing events identifies a novel biomarker of TDP-43 pathology

    Fine-tuned OpenSpliceAI, a specialized splice site prediction model, outperformed a splicing head on AlphaGenome's frozen backbone for detecting TDP-43 cryptic exons, though fine-tuning all AlphaGenome weights enabled generalization to held-out data. Genome-wide application of fine-tuned OpenSpliceAI identified new cryptic splicing events in CLDN11 and ARHGAP23, validated in FTLD-TDP brain tissue. The CLDN11 cryptic exon is predicted to generate a neoepitope that could serve as an oligodendrocyte-specific biomarker of TDP-43 pathology.

  4. biospace.com54

    TRexBio, Retension raise $161.7M in IPOs; City, Iambic set proposed offering terms

    TRexBio raised $116.7 million in its Nasdaq IPO, selling 8.3 million shares at $14 to fund TRB-061, a fusion protein in early-stage development for atopic dermatitis with a Phase 1b topline readout expected mid-2027. Retension Pharmaceuticals raised $45 million at $12 per share to advance RTN-001, a PDE-5 blocker for hypertension, through Phase 2b and into Phase 3. City Therapeutics expects about $158.3 million and Iambic Therapeutics about $135 million from upcoming offerings.

    Why it matters: Four biotech IPOs in one week signal a reopening of the public markets for drug developers.

  6. nature.com52

    Will AI scoop your science? Some researchers see a gloomy future

    Scientists are warning that autonomous AI agents may scoop their research after two recent cases. OpenAI announced on 8 September that its agents had solved the Navier–Stokes problem one day after mathematician Tristan Buckmaster posted progress on it, and Anthropic reported that its Claude LLM agents found CRISPR-like repeated DNA patterns in viruses that PhD student Mario Rodríguez Mestre had been studying unpublished for years. Both companies denied training on user data, and researchers including Sandra Laurentino and Samuel Mehr are now restricting what they upload to commercial AI tools.

  7. nature.com47

    Phylogeny-informed genomic prediction with GPN-Star

    GPN-Star, a phylogeny-informed genomic language model, predicts functional genetic variants contributing to inherited traits by learning evolutionary constraints from multispecies whole-genome alignments and the corresponding phylogeny. Trained on vertebrate, mammalian and primate alignments, it predicts coding and non-coding variant effects on human disease susceptibility and enhances enrichment of complex trait heritability. Versions trained on five model organisms, from Mus musculus to Arabidopsis thaliana, show the approach generalizes across species.

Oct 8Thu
  1. biospace.com66

    Big Pharma–backed Matchpoint scores $150M series B, advancing autoimmune asset

    Preclinical Matchpoint Therapeutics has raised a $150 million series B co-led by Nextech Invest and Norwest, with participation from Sanofi Ventures, Atlas Venture, Access Biotechnology and the T1D Fund. The money will advance MPT-062, an oral T cell modulator for autoimmune and inflammatory conditions, into human trials starting next year. Matchpoint also has a July 2025 deal with Novartis, which paid $60 million upfront for its ACE platform and could owe up to $1 billion in milestones plus royalties.

    Why it matters: A large series B funds a first-in-class oral autoimmune program before it reaches human trials, against a market trend favouring later-stage assets.

  2. Nat Med46

    AI-assisted scoping review of code sharing in clinical prediction model research.

    Only 482 of 3,967 articles citing TRIPOD or TRIPOD+AI (12.2%) included code-sharing statements, according to a large-language-model-assisted scoping review that screened articles and assessed repositories against 14 reproducibility-related features. Sharing varied by journal and country, and repositories showed substantial heterogeneity. The results inform TRIPOD-Code, a reporting guideline for code availability and reproducibility.

Oct 7Wed
  1. Sci Adv58

    Deep learning extracts MoA-specific signatures from high-throughput images of chemically and genetically perturbed Corynebacteria.

    A deep learning model classifies modes of action directly from high-throughput images of drug-treated Corynebacterium glutamicum, a surrogate for Mycobacterium tuberculosis. The model classifies MoAs of established antibiotics, recognizes MoAs of previously unseen antibiotics, and clusters inhibitors with a previously unseen MoA apart from reference drugs, enabling detection of novel MoAs. It also links drug and mutant images targeting similar pathways, supporting mutant-based target prediction for compounds with novel MoAs.

  2. Nat Biotechnol63

    Decoding systemic vascular health and hypertensive disorders in pregnancy through retinal imaging and Visionary AI.

    Visionary AI, a platform combining retinal imaging with graph-based vascular modeling, predicted hypertensive disorders of pregnancy before clinical onset. In a prospective development cohort of 1,267 pregnancies it reached AUC 0.91 and average precision 0.81 for preeclampsia, and a stability-optimized model transferred without retraining to external validation at AUC 0.81 and AP 0.68, outperforming clinical and deep learning benchmarks. The finding positions the maternal retina as a minimally invasive biosensor for first-trimester risk stratification.

    Why it matters: First-trimester preeclampsia prediction from retinal images outperformed clinical and deep learning benchmarks in external validation.

  3. nature.com46

    Why sound, strong science alone isn’t enough in pandemic preparedness

    Polina Brangel, research-and-development data-innovation lead at the Coalition for Epidemic Preparedness Innovations (CEPI) in Oslo, is using AI to prepare for Disease X, the WHO term for a hypothetical pandemic pathogen. CEPI's programme works across the 25 known families of virus that infect humans, using simulations and AI to predict mutations and design vaccine candidates. She is also developing CEPI's Pandemic Preparedness Engine, a chatbot-like tool that would let researchers input a pathogen's genomic sequence and receive suggestions for vaccine antigens.

  4. Cell56

    Glioma margins gain a third dimension before surgery ends.

    ULTRA generates cell-resolved three-dimensional virtual histology of intact glioma specimens in about 30 minutes by combining rapid tissue clearing, single-channel stimulated Raman scattering microscopy, and AI. The speed makes it possible to expose depth-dependent infiltration before surgery ends, giving glioma margins a third dimension during the operation. Liu et al. present the method in Cell.

Oct 6Tue
  1. biospace.com46

    Compliance concerns hinder rollout of AI in drug manufacturing: Survey

    Fewer than 10% of drugmakers are scaling AI in clinical or commercial manufacturing, according to a CRB survey of more than 400 R&D and manufacturing professionals for its 2026 Horizons report. Use is concentrated in nonregulated areas such as administrative workflows (56%) and drug discovery (45%), with accuracy and quality concerns and insufficient validation standards cited as the main barriers. CRB found that 31% of respondents plan to use AI in regulated manufacturing within three years, and another 31% are considering it without a formal plan.

    Why it matters: Shows regulatory and validation concerns are blocking AI adoption in drug manufacturing, with an FDA warning letter already issued.

  2. nature.com46

    Cells, who needs them? Biochemists turn to proteins in a tube

    Cell-free protein-expression systems, built from cell lysates or purified components (PURE), let protein designers screen thousands of options without growing cells, cutting design-build-test-learn cycles from days or weeks to hours. Applications include biosensors such as ROSALIND, which detected copper and zinc in water supplies after a 2018 California wildfire, and a machine-learning-guided lead sensor detecting 5.7 parts per billion. Costs remain high, with litre-scale reagents above US$4,000 and most studies making under two grams of protein.

  3. nature.com65

    Decoding systemic vascular health and hypertensive disorders in pregnancy through retinal imaging and Visionary AI

    Hypertensive disorders of pregnancy affect 2–15% of pregnancies and are usually diagnosed after 20 weeks of gestation, once hypertension, proteinuria or organ dysfunction appear, even though low-dose aspirin started before 16 weeks can reduce risk of preterm delivery with preeclampsia by up to 62%. The paper develops retinal imaging with AI to detect the systemic and placental vascular changes that precede symptoms, building on evidence that women with preeclampsia show retinal arteriolar narrowing, increased tortuosity and reduced capillary density before diagnosis.

    Why it matters: It tests a noninvasive retinal imaging route to early preeclampsia risk identification, which current late diagnostics fail to provide.

  5. nature.com60

    MAGIC: an international network for evaluating generative artificial intelligence in global health

    MAGIC is an international network, described in Nature Medicine, for evaluating generative artificial intelligence in global health. Its contributors are based at institutions including Duke-NUS Medical School, Weill Cornell Medicine, the University of Tokyo, the WHO Regional Office for South-East Asia, the International Telecommunication Union, Harvard Medical School and the University of Ghana Medical School. The effort spans biomedical data science, computational biology, clinical medicine and health policy across Asia, Africa, Europe and North America.

    Why it matters: A new international network sets out how generative AI will be evaluated for use in global health.

  6. biorxiv.org51

    Discovering Latent Scientific Concepts through Discrete Representation Learning

    Kodiak, a discrete representation learning architecture, learns latent scientific concepts from small, curated scientific imaging datasets through Sinkhorn-balanced codebook assignments, avoiding the instability of continuous self-supervised adaptation where teacher representations drift. Across five scientific imaging domains, including multiplex immunofluorescence for pancreatic cancer pathology and scanning electron microscopy, it improves representation quality under both foundation-model adaptation and training from scratch. The learned concepts are interpretable, reusable, and produce spatially coherent concept maps supporting cell phenotyping, spatial niche discovery, tissue organization, and materials characterization.

CBACBA Frontier Insights (CFI)Chinese Biopharmaceutical Association (CBA)www.cba-usa.org