BioLLM Wiki โ€บ Clinical & Outcomes

Machine Learning & AI

These papers apply machine learning and artificial intelligence to single-cell RNA-sequencing, small-molecule, pathology image and language, and electronic health record data for prognostic modeling, disease and toxicity prediction, risk assessment, pathology question answering, and clinical documentation. They examine macrophage-related prognosis and immunotherapy efficacy in lung adenocarcinoma, molecular and immune features of type 2 diabetes, small-molecule toxicity prediction for drug development, emergency-department gastrointestinal-bleeding risk stratification, a vision-language assistant for pathology, and ambient AI documentation in outpatient orthopaedic care.

Papers
10,144
Jan 2024 โ€“ Aug 2026
2024 share
7.52%
of the month's papers, on average
2025 share
8.86%
of the month's papers, on average
2026 share
9.40%
of the month's papers, on average
Change
โ–ฒ +25.0%
2024 โ†’ 2026, relative

Share of papers by month

Peak 10.65% in Aug 2026. Month-by-month values and sub-terms are on Keyword Trends.

Jan 2024Aug 2026

Molecular biomarkers and prognosis

Machine-learning models are combined with transcriptomics and single-cell data to identify disease-associated genes and patient subgroups. In lung adenocarcinoma, single-cell analysis characterized macrophage states and their evolutionary trajectories, while TCGA data supported a macrophage-gene prognostic model; in type 2 diabetes, 12 algorithms were used to model metal-ion-transport genes and immune-cell infiltration; and in Alzheimer disease, machine learning identified five candidate biomarkers that were subsequently examined with single-cell and immune-infiltration analyses (PMID 39850883; PMID 39926598; PMID 40190307). Similar workflows selected MORN3 and LLGL2 for latent tuberculosis diagnosis, integrated cuproptosis- and ferroptosis-related genes for melanoma prognosis and treatment-response prediction, and identified platelet-related genes associated with ulcerative colitis immune-cell populations (PMID 39054612; PMID 41613147; PMID 42668808).

๐Ÿ“„ Frontiers in immunology โ†—๐Ÿ“„ Frontiers in immunology โ†—๐Ÿ“„ Archives of medical science : AMS โ†—๐Ÿ“„ Annals of medicine โ†—๐Ÿ“„ Frontiers in immunology โ†—๐Ÿ“„ PeerJ โ†—

Medical imaging and pathology

Deep-learning and machine-learning systems in these papers analyze MRI, CT, endoscopic, pathology-slide, retinal, and mammographic images. Examples include automatic Cobb-angle measurement from segmented lumbar MRI, total bleeding-volume quantification on admission CT in aneurysmal subarachnoid hemorrhage, temporal-regional graph analysis of knee MRI for osteoarthritis progression, and AI-assisted detection of esophageal squamous cell carcinoma in a randomized clinical endoscopy trial (PMID 38383922; PMID 38552240; PMID 38686707; PMID 39317205). Other models used fundus images to address cross-device domain shifts in diabetic-retinopathy grading, whole-slide hematoxylin-eosin images to predict uveal-melanoma molecular subclasses, and mammographic radiomics or fused radiomics/deep-learning outputs to distinguish difficult normal cases or benign BI-RADS 4 calcifications (PMID 38701588; PMID 38952117; PMID 39407048; PMID 39633212).

๐Ÿ“„ European radiology โ†—๐Ÿ“„ Journal of neurosurgery โ†—๐Ÿ“„ Journal of magnetic resonance imaging : JMRI โ†—๐Ÿ“„ Endoscopy โ†—๐Ÿ“„ Computers in biology and medicine โ†—๐Ÿ“„ Histopathology โ†—๐Ÿ“„ Journal of imaging informatics in medicine โ†—๐Ÿ“„ Journal of imaging informatics in medicine โ†—

Computational drug discovery

Machine learning is used to prioritize compounds, predict toxicity, and balance pharmacological activity with drug-like properties. ToxMPNN uses a message-passing neural network trained across 27 toxicity endpoints and seven toxicity classes, whereas a clinical-adverse-event model combined predicted on- and off-target IC50 values for 759 approved drugs with tissue-specific protein expression (PMID 38409892; PMID 39095534). For target discovery, QSAR classifiers screened DrugBank compounds for PI3Kฮฑ inhibition, and integrated ML, docking, and molecular-dynamics workflows evaluated SARS-CoV-2 Mpro and MMP-13 inhibitors, including the trade-off between Mpro affinity and pharmacokinetic properties (PMID 38507171; PMID 40733553; PMID 41037200). A de novo design workflow for A2A adenosine-receptor antagonists incorporated receptor interactions, blood-brain-barrier permeability, and multiparameter optimization, producing compounds with predicted CNS permeability and nanomolar affinity in one design run (PMID 42319230).

๐Ÿ“„ Journal of applied toxicology : JAT โ†—๐Ÿ“„ Pharmaceutical research โ†—๐Ÿ“„ Applied biochemistry and biotechnology โ†—๐Ÿ“„ Viruses โ†—๐Ÿ“„ Molecular diversity โ†—๐Ÿ“„ Journal of medicinal chemistry โ†—

Clinical risk and treatment prediction

Several studies train models on routinely collected clinical data to estimate complications, treatment response, or healthcare utilization. Causal survival forests within an emulated target trial associated SSRI treatment with a higher risk of antibiotic-resistant infection, while EHR-based models were externally validated for gastrointestinal-bleeding outcomes and logistic regression, LASSO, random forest, and XGBoost models predicted acute opioid-use-disorder events among Medicaid enrollees with performance assessed across racial groups (PMID 40417572; PMID 38971198; PMID 40302110). Other applications predicted postoperative infarction after moyamoya revascularization, chemotherapy response in pancreatic cancer from plasma-exosomal microRNAs, and radiation pneumonitis in lung-cancer patients with interstitial lung disease using clinical, dosimetric, and inflammatory variables (PMID 38579355; PMID 39278808; PMID 41755502).

๐Ÿ“„ AMIA ... Annual Symposium proceedings. AMIA Symposium โ†—๐Ÿ“„ Gastroenterology โ†—๐Ÿ“„ Addiction (Abingdon, England) โ†—๐Ÿ“„ Journal of neurosurgery โ†—๐Ÿ“„ Pancreatology : official journal of the International Association of Pancreatology (IAP) ... [et al.] โ†—๐Ÿ“„ Annals of medicine โ†—

Generative AI and clinical workflows

Generative and language-based systems are evaluated as assistants for pathology, radiology-report review, documentation, education, and patient communication. PathChat combines a pathology vision encoder with a pretrained large language model and was fine-tuned on more than 456,000 visual-language instructions; GPT-4 and Gemini were compared for matching findings across serial abdominal CT reports; and an ambient AI documentation system was assessed using note length, documentation time, after-hours work, and same-day closure metrics (PMID 38866050; PMID 39658474; PMID 40982636). Other studies compared ChatGPT-based and human simplification of orthopaedic education materials for readability and fidelity, and compared several AI models with oculoplastic surgeons when answering ocular-dyskinesia questions (PMID 41747019; PMID 40865167). Governance papers emphasize implementation barriers, human oversight, transparency, privacy, and fairness when translating healthcare AI from research into practice (PMID 39470044; PMID 41795494).

๐Ÿ“„ Nature โ†—๐Ÿ“„ Academic radiology โ†—๐Ÿ“„ The Journal of the American Academy of Orthopaedic Surgeons โ†—๐Ÿ“„ The Journal of bone and joint surgery. American volume โ†—๐Ÿ“„ Ophthalmic plastic and reconstructive surgery โ†—๐Ÿ“„ Healthcare management forum โ†—๐Ÿ“„ International journal of medical informatics โ†—

Written by gpt-5.6-luna from 96 of this keyword's papers, a few from every whole month through Aug 2026; every paragraph was checked against the abstracts it cites. A paper counts for the keyword when its title, abstract, keywords or MeSH headings match the keyword's terms. Not medical advice; see the Disclaimer.