Skip to content

#AI制药

Today 0 items
Oct 15Thu
  1. 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.

  2. 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.

  3. 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.

  4. 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.

Oct 14Wed
  1. 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 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.

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. 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.

Oct 6Tue
  1. 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.

  2. biorxiv.org49

    Distinct sequence grammars of nucleosomal and linker DNA shape mutational and methylation landscapes in cancer

    DeepND, an interpretable deep learning framework, distinguishes nucleosomal from inter-nucleosomal DNA and identifies 210 sequence motifs of 2-16 bp length that are significantly enriched or depleted in nucleosomal DNA. Forty-four of these motifs show cancer-type-specific enrichment or depletion of somatic mutations, and six are associated with differential DNA methylation levels. In lung and endometrial cancers, mutations show a significant ~10-bp periodic enrichment pattern across distinct minor-groove orientations, linking mutational and repair processes to nucleosome rotational positioning.

  3. biorxiv.org49

    Mechanistic Decoding of Pharmacodynamic Intratumoral Heterogeneity: A Pan-Cancer Regulatory Framework

    A pan-cancer analysis links single-cell transcriptomes from 1,279 samples across 24 cancer types with drug-response profiles for 1,218 compounds, defining 425 hub genes associated with predicted treatment-response heterogeneity. A multi-omics case study in lung adenocarcinoma identifies four regulatory classes of candidate associations, including 3 transcription factors, 41 genomic alterations, 5 ligand-receptor pairs and 8 splicing genes, converging on 6 core tumor-intrinsic pathways. An online platform enables state-resolved drug-response analysis and candidate-gene exploration.

Sep 29Tue
Sep 28Mon
  1. nature.com56

    AI-guided optimization for thermostable mRNA vaccines

    AGENT, an AI-driven framework combining high-throughput experimentation with Bayesian optimization, engineers thermostable solid-state mRNA vaccines. The formulations retained full bioactivity after 2 months at 37 °C and elicited immune responses non-inferior to fresh vaccines. Microneedle patches enabled cold-chain-free delivery in rodents and nonhuman primates.

    Why it matters: Cold-chain-free mRNA vaccines with retained bioactivity could change vaccine distribution logistics.

  2. nature.com42

    Accelerated discovery of thermostable mRNA–lipid nanoparticle vaccines using data-efficient AI

    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.

    Why it matters: It shows an AI-driven route to thermostable mRNA-LNP vaccines, a known cold-chain bottleneck.

Sep 24Thu
  1. biorxiv.org45

    CMAC: A deep learning framework for absolute cardiomyocyte transcriptional age estimation

    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.

    Why it matters: It provides the first unified way to quantify absolute cardiomyocyte maturation age and deviations from normal development.

Sep 23Wed
  1. Nature Medicine41

    AI-based characterization of Alzheimer’s disease phenotypes from population-scale single-cell data

    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.

    Why it matters: Population-scale single-cell data plus a graph neural network offer an AI route to define Alzheimer's phenotypes.

Sep 22Tue
  1. Nature Medicine52

    Large-scale esophageal cancer screening through noncontrast computed tomography and artificial intelligence

    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.

    Why it matters: It points to CT-based AI screening as a route to wider esophageal cancer detection.

  2. medrxiv.org54

    Genomic foundation models extend clinicopathologic and transcriptomic prognostication in soft tissue sarcoma

    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.

    Why it matters: It shows genomic foundation model features improve established sarcoma prognostication across cohorts and survival endpoints.

Sep 21Mon
  1. medrxiv.org54

    A Causal Multi-modal AI Model Stratifies Residual Risk and Identifies Candidates for Treatment Escalation in Node-Positive HR+/HER2- Early Breast Cancer

    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.

    Why it matters: An AI model identifies which high-risk HR+/HER2- breast cancer patients can avoid escalation and who may benefit from everolimus.

Sep 18Fri
  1. nature.com47

    A universal visual foundation model for computational cytopathology

    CROWN is a universal visual foundation model for computational cytopathology, benchmarked on more than 200 real-world cytology tasks across cohorts, including lymph node metastasis and cervical screening datasets. It gives cytopathology a single pretrained model that can be evaluated across many clinical cytology settings rather than task-specific systems.

    Why it matters: A single foundation model covering over 200 real-world cytology tasks could standardize computational cytopathology workflows.

Sep 17Thu
  1. medrxiv.org61

    Population-Scale Precision Safety in Oncology Reveals Clinical and Genetic Determinants of Systemic Therapy Toxicity

    A large language model-based resource, MSK-Tox, captured the incidence, timing, and grade of anti-cancer therapy toxicity across more than 50,000 patients. Analysis of six representative adverse events showed toxicity risk varies by cancer type, regimen, and clinical context, is predictable from pretreatment clinical features, and has two germline modes: an organ-intrinsic mode, exemplified by a regulatory variant near FOXE1 linked to hypothyroidism, and an immune-mediated mode in which HLA-DRB1*15 drives adrenal insufficiency after immune checkpoint inhibition. The work supports pretreatment risk assessment for harm on the same principles used to predict benefit.

    Why it matters: It shows germline and clinical predictors can identify who will be harmed by cancer therapy before treatment starts.

  2. Nat Med50

    On-premise medical AI agents for reliable clinical decision-making.

    A fully on-premise clinical AI agent, combining local operational control with a multi-perspective reliability framework, achieved 90.04% accuracy on a seven-disease MIMIC-IV-derived benchmark and 83.8% on a four-disease task, approaching a cloud baseline. Behavioral consistency best discriminated correct diagnoses (AUC = 0.860; 0.875 under stress testing); at a 0.90 threshold, 49.4% of cases were retained at 98.9% accuracy. The framework lets institutions route a lower-risk subset to autonomy and defer the rest for review.

    Why it matters: It shows an on-premise agent whose decision-time reliability signals select which cases can run autonomously at high accuracy.

  3. Nat Med64

    Prediction of maternal and infant outcomes from longitudinal electronic health records with a Mother-Child AI agent.

    MoChiAgent, an LLM-based clinical assistant built on the MoChiFormer model, forecasts maternal and infant diseases from routine longitudinal EHR data, using 4,401,599 clinical visits for development and external validation in 263,452 maternal and 23,192 infant visits. It identified placental abruption and premature rupture of membranes at AUROCs of 0.89 and preterm labour at 0.91, and linked specific maternal clusters to neonatal jaundice (HR 2.81) and haematological diseases (HR 2.83) in infants. Pairing maternal gestational records with infant records improved prediction of infant conditions such as chromosomal abnormalities and respiratory disorders.

    Why it matters: Large-scale external validation shows routine EHR data can replace costly tests for maternal and infant risk prediction.

Sep 14Mon
  1. Nature69

    Creating bottom-up RNA transfer vehicles from synthetic protein assemblies.

    AI-designed protein assemblies combined with natural protein domains yielded more than 100 bottom-up RNA transfer vehicles with unique sizes and shapes, surpassing the RNA transfer efficiency of widely used delivery vehicles by several orders of magnitude. Tropism was programmed with computationally designed peptide binders, enabling delivery of therapeutically relevant cargo RNAs across cellular models. One vehicle's in vivo biodistribution was mapped in a mouse at near-single-cell resolution, confirmed as safe, and used for a gene-editing treatment strategy for Duchenne muscular dystrophy in patient-derived cells and a pig.

    Why it matters: Generative AI-designed protein assemblies outperform established RNA delivery vehicles and work in vivo.

  2. Nature54

    Predicting genome-wide functional constraints with GPN-Star.

    GPN-Star, a phylogeny-aware genomic language model that uses whole-genome alignments and species trees, outperforms previous methods across variant effect prediction tasks in coding and non-coding human genome regions. Trained on vertebrate, mammal and primate alignments, it improves prioritization of pathogenic and fine-mapped GWAS variants, yields strong complex trait heritability enrichments, and improves rare variant association testing power. Versions trained for five model organisms, from mouse to Arabidopsis, show the framework generalizes.

    Why it matters: A phylogeny-aware language model that beats prior methods on variant effect prediction and human genetics applications across species.

Sep 13Sun
  1. nature.com49

    Clinical usability of an explainable AI decision support tool and evaluation of multimodal models in NSCLC

    A multimodal explainable AI model predicted immunotherapy outcomes in non-small cell lung cancer better than established biomarkers, in a large international real-world study. The tool also improved physician decision-making. It offers an explainable alternative to current biomarkers for guiding immunotherapy choices.

    Why it matters: A real-world study shows explainable AI beating established biomarkers and aiding physician decisions in NSCLC immunotherapy.

Sep 12Sat
  1. medrxiv.org45

    AI detects a distributed blood metabolomic Systemotype associated with early stage ovarian cancer

    A machine-learning classifier detected early-stage ovarian cancer from untargeted metabolomic profiles with cross-validated sensitivity and specificity of 99.0% and 99.8% in serum (59 cancer, 32 controls) and 97.7% and 99.7% in plasma (63 cancer, 20 controls). Both cohorts shared 239 concordantly altered features across lipid, amino-acid, steroid, central-carbon, and redox metabolism, supporting a distributed metabolic 'Systemotype' detectable in blood.

    Why it matters: Reported sensitivities and specificities far exceed those of existing ovarian cancer blood biomarkers used for screening.

Sep 11Fri
  1. Nat Med53

    Clinical usability of an explainable AI decision support tool and evaluation of multimodal models in NSCLC.

    In the I3LUNG study of 2,396 patients with non-small cell lung cancer, machine learning and deep learning models using clinical and blood data predicted immunotherapy outcomes with AUC up to 0.77 in the test set and surpassed PD-L1, ECOG performance status, NLR, LDH and LIPI scores. External validation gave AUC of 0.55 to 0.72. An explainable AI tool improved predictions by lung expert and nonexpert physicians, while multimodal fusion added no consistent benefit over clinical and blood data alone. Prospective validation in more than 2,000 patients is underway.

    Why it matters: It tests AI-based immunotherapy decision support against current PD-L1-based selection in a large real-world cohort.

  2. Cancer Cell52

    Ensemble learning of pathology foundation models for precision oncology.

    ELF integrates five pretrained pathology foundation models into unified slide-level representations, trained on 53,699 whole-slide images across 20 anatomical sites. It outperformed its constituent and slide-level foundation models on disease classification, biomarker detection, and anticancer and immunotherapy response prediction across multiple cancer types. Its architecture allows data-efficient downstream evaluation, including settings with limited data such as therapeutic response prediction.

    Why it matters: Ensemble learning of pathology foundation models improved slide-level performance across precision oncology tasks.

Sep 7Mon
  1. nature.com66

    Integration of proteomic aging clocks in a phase 2a clinical trial supports simultaneous geroprotective assessment

    Six proteomic aging clocks, applied to longitudinal serum proteomics from 42 participants with idiopathic pulmonary fibrosis in a phase 2a trial of rentosertib, an AI-designed TNIK inhibitor, all showed a statistically significant decrease in the pace of aging. Chronological-age-trained clocks tracked aging rate modulation more accurately than mortality-trained clocks, supporting multi-clock proteomic assessment as a trial endpoint for interventions targeting aging biology.

    Why it matters: Shows multiple proteomic aging clocks converging on slowed aging in a controlled human trial of a geroprotective-targeted drug.

Sep 3Thu
  1. Nat Biotechnol62

    Adaptive model-guided protein evolution with sparse data optimizes compact eukaryotic genome editors.

    A model-guided engineering strategy for sparse data, EvoMax, produced a compact eukaryotic Fanzor2 nuclease variant, FanzMAX v3-hLa, that reaches up to 97% editing efficiency at the best-performing endogenous locus and a mean of ~33% across 19 endogenous loci. It outperformed enNlovFz2 and enCnCas12f1 by more than 2.6-fold and edited hPCSK9 in humanized mice. EvoMax combines Gaussian process regression, protein language models and inverse folding with iterative experimental profiling.

    Why it matters: It delivers a compact programmable nuclease with substantially higher mammalian editing efficiency than established compact editors.

Sep 1Tue
  1. Cell59

    Viral protease-initiated lytic cell death as a universal antiviral mRNA therapy.

    Gasdermin-D was engineered to carry viral protease-specific cleavage motifs, producing VID activators that trigger lytic cell death only in infected cells. Delivered as LNP-encapsulated mRNA in a hepatitis A virus model, they abolished viral replication and shedding in vivo and eased liver injury via a kill-and-alert mechanism, with activity also shown against Zika virus and SARS-CoV-2. Generative AI-designed cleavage motifs for the SARS-CoV-2 main protease improved antiviral potency.

    Why it matters: It turns viral protease activity into a broadly applicable mRNA kill-and-alert antiviral strategy.

Aug 30Sun
  1. Nat Biotechnol72

    A blinded, prospective benchmark of in silico antibody discovery anchored to experimental affinity and developability.

    A blinded, prospective benchmark (AIntibody) tested 511 AI-designed or predicted antibodies from 29 organizations across three tasks: affinity maturation, affinity ranking within HCDR3 clusters, and CDR design outside selection outputs. Several groups produced developable antibodies with affinities <100 pM, but successes did not transfer across tasks; affinity ranking was worse than random clone picking except for one model, and out-of-library design often failed to beat standard selections. The results show AI can optimize antibodies in defined, biologically grounded regimes while exposing gaps in affinity prediction, library-inspired design, and cross-task generalization.

    Why it matters: First blinded prospective benchmark separating real AI antibody design gains from hype across three tasks.

Aug 29Sat
  1. JAMA44

    Electrocardiogram-Based Deep Learning to Prioritize Testing for Transthyretin Amyloid Cardiomyopathy.

    A locally deployable AI model reading routine ECG images identified transthyretin amyloid cardiomyopathy with an AUROC of 0.84 in internal validation (44 123 patients) and 0.78 to 0.89 across 5 external cohorts, holding up in screening settings of older Black and Hispanic adults with heart failure and adults with prior carpal tunnel surgery. Adding AI-enabled echocardiography after a positive AI-ECG raised positive predictive value from 0.24 to 0.66 while sensitivity fell from 0.84 to 0.68. The approach offers an accessible way to prioritize patients for confirmatory imaging.

Aug 24Mon
  1. nature.com70

    Adaptive model-guided protein evolution with sparse data optimizes compact eukaryotic genome editors

    A model-guided engineering pipeline (EvoMax) combining transfer-learning Gaussian process regression, ESM-2 and ESM-IF optimized Fanzor2 nucleases, compact eukaryotic genome editors, using sparse experimental data. Screening over 1,600 Fz2-like sequences yielded active orthologs including NaloFz2, and the resulting FanzMAX v3-hLa system reached up to 97% editing efficiency (mean ~33% across 19 loci) in bulk mammalian cells, roughly tenfold above baseline, with expanded TAM compatibility. In vivo editing of hPCSK9 in mice was achieved with single-AAV delivery.

    Why it matters: It gives laboratories a compact single-AAV-deliverable nuclease with near-homogeneous editing in bulk mammalian cells.

Aug 22Sat
  1. Cancer Cell65

    Spatial biomarker discovery via interpretable semantic learning in histopathology.

    PathPrism, an interpretable AI framework, encodes tissue architecture into pathologically informed spatial features to model prognosis, molecular alterations, and therapy response. Applied to 7,000 patients with colorectal cancer across 11 cohorts, it uncovered hundreds of biomarkers predictive of survival, MSI, BRAF, and TP53 mutations, and stratified chemotherapy benefit in stage II/III disease. It also uses large language models to generate hypotheses and includes VirtualWSI for semantic perturbation.

    Why it matters: It offers an interpretable alternative to black-box models for discovering spatial biomarkers at cohort scale.

Aug 19Wed
CBACBA Frontier Insights (CFI)Chinese Biopharmaceutical Association (CBA)www.cba-usa.org