← bioTriage-Open

Machine Learning for Biological Data

Deep learning, protein language models, GNNs, and foundation models applied to biological sequence and structure.

30 papers curated · High: 1 · Medium: 21 · Low: 8 · Updated: 2026-11-19

Renal failure 2026-12-01 Medium

A machine learning-based predictive tool for complete renal response in lupus nephritis leverages the 3-month treatment milestone.

This prospective cohort study of 400 active lupus nephritis patients used machine learning to predict complete renal response at 12 months, leveraging predictors measured at baseline and the 3-month treatment milestone. The authors developed interpretable ML models showing that early treatment response data can forecast longer-term renal outcomes, potentially enabling earlier clinical intervention.

machine-learningclinicalimmunology
Journal of Korean medical science 2026-11-16 Medium

Machine Learning Based Plasma Proteomic Analysis Identifies Disease Differentiating Biomarkers in Atopic Dermatitis.

This study applied machine learning to plasma proteomic data to find proteins that distinguish atopic dermatitis from ulcerative colitis, two inflammatory diseases that both involve aberrant Th2 immune activity but affect different tissues. The analysis identified candidate blood-based biomarkers that differentiate the two conditions, offering insight into their distinct disease mechanisms.

proteomicsmachine-learningclinical
Journal of Korean medical science 2026-11-16 Medium

A Deep Learning-Based Prediction Model for Mortality in Severe COVID-19 Patients Requiring Intensive Care Unit Admission: A Multicenter Registry Analysis.

This multicenter registry study developed a deep learning model to predict mortality among critically ill COVID-19 patients admitted to the ICU, using a deliberately minimal set of readily available clinical variables to enable rapid risk stratification. The work addresses a gap in AI-based mortality prediction, which has focused less on the ICU population, and emphasizes practical clinical deployability. The model's performance and validation across centers are the key contributions.

machine-learningclinicaldeep-learning
Cancer science 2026-11-16 Medium

MARiO: Predicting Cancer Variant Pathogenicity by Integrating In Silico Evaluation and Patient-Level Mutational Contexts.

MARiO is a machine-learning model that predicts the pathogenicity of missense cancer variants by combining variant-level features with patient-level mutational context from comprehensive genomic profiling. Unlike most existing tools that rely only on variant-level information, MARiO incorporates clinical context to improve pathogenicity assessment, particularly for variants lacking established consensus.

machine-learningcancergenomics
JMIR medical informatics 2026-11-13 Medium

Deep-Learning Analysis of Peripheral Blood Smear for Prediction of Acute Inflammatory States and Chronic Comorbidities: Development and Internal Validation Study.

This study develops and internally validates a deep-learning model that analyzes peripheral blood smear images to predict acute inflammatory states and chronic comorbidities. The work suggests that morphological features of immune cells captured on routine blood smears contain systemic pathophysiological information beyond standard cell counts. Internal validation indicates the approach is feasible, though external validation is still needed.

deep-learningclinicalmethodology
Cell 2026-11-12 High

Single-cell analysis of chromatin state and transcriptome in the human basal ganglia.

This study presents a single-cell multi-omic atlas of histone modifications and transcriptomes across eight human basal ganglia regions, revealing cell-type- and region-specific regulatory elements and gene regulatory networks. The authors identify combinatorial homeobox transcription factor expression as a key driver of cell identity in this brain region.

epigenomicstranscriptomicssingle-cell
Cancer science 2026-11-10 Medium

Multimodal Integration Identifies C16orf74 as an EMT Regulator With Prognostic Relevance in HNSCC.

This study integrates single-cell and bulk multi-omics data to dissect malignant epithelial heterogeneity in head and neck squamous cell carcinoma, identifying C16orf74 as a novel regulator of epithelial-mesenchymal transition (EMT). The authors show that C16orf74 expression is associated with EMT-associated malignant states and has prognostic relevance in HNSCC patients. The work nominates a previously uncharacterized gene as a potential biomarker and therapeutic target in this aggressive cancer.

single-celltranscriptomicscancer
Advanced science (Weinheim, Baden-Wurttemberg, Germany) 2026-11-10 Medium

EZpred: Improving Deep Learning-Based Enzyme Function Prediction Using Unlabeled Sequence Homologs.

EZpred is a deep learning method that improves enzyme function prediction (EC number assignment) by incorporating features from unlabeled sequence homologs, analogous to how homolog-aware structure predictors like AlphaFold outperform single-sequence methods. This addresses a gap where most deep learning function predictors ignore evolutionary information from homologous sequences. The approach likely boosts accuracy by leveraging unlabeled homolog data without requiring additional labeled examples.

machine-learningprotein-structuremethodology
Molecular biomedicine 2026-11-09 Medium

Preoperative discrimination of malignant and benign ovarian masses using cfDNA fragmentomics and serum biomarkers.

This multicenter study developed a diagnostic model combining cfDNA fragmentomic features (fragment size, nucleosome footprints, copy-number variations, repeat elements) from low-pass whole-genome sequencing with serum biomarkers to distinguish malignant from benign ovarian masses. The stacked machine-learning model was trained and validated across 743 participants in multiple cohorts, addressing the clinical challenge of ambiguous preoperative diagnosis when standard serum markers are uninformative.

genomicsmachine-learningclinical
Inflammatory bowel diseases 2026-11-09 Medium

Interpretable machine learning applied to high-dimensional salivary proteomics in pediatric inflammatory bowel diseases.

This proof-of-concept study applies interpretable machine learning to high-dimensional salivary proteomics to identify a minimal protein signature that can classify pediatric inflammatory bowel disease subtypes (Crohn's disease, ulcerative colitis, and IBD-unclassified). The approach aims to provide a non-invasive alternative to current invasive and costly diagnostic methods, potentially enabling earlier diagnosis and better longitudinal monitoring.

proteomicsmachine-learningclinical
Journal of chemical information and modeling 2026-11-09 Medium

FlexENN: A Graph Neural Network for Binding Energy Prediction of Globular and Intrinsically Disordered Proteins.

FlexENN is a graph neural network framework for predicting binding energies of protein complexes, with separate models for folded complexes (FlexENN-F) and complexes involving intrinsically disordered proteins (FlexENN-I). It builds local interface graphs to capture binding-relevant features, addressing the difficulty of predicting binding energies for disordered interfaces that lack defined geometry.

machine-learningprotein-structuremethodology
Geburtshilfe und Frauenheilkunde 2026-11-01 Medium

Artificial Intelligence in Obstetrics and Prenatal Medicine: Current Evidence, Clinical Validation and Implementation.

This review surveys the current evidence for AI applications in obstetrics and prenatal medicine, covering prenatal ultrasound, fetal echocardiography, fetal MRI, genomic screening, risk prediction, and clinical decision support. It emphasizes that while AI tools are increasingly entering prenatal care, clinical validation and implementation requirements remain key considerations before routine adoption.

machine-learningclinicalreview
PLoS computational biology 2026-11-01 Medium

Estimating rates of SARS-CoV-2 lineage spread using graph spectra from genetic sequence data.

The authors present a new method that combines graph theory and machine learning to identify SARS-CoV-2 lineages undergoing rapid population growth from large volumes of genetic sequence data. By representing distances between small subsets of sequences as graph Laplacians and extracting spectral features, the approach can flag emerging lineages in real time without analyzing the full dataset. This offers a scalable way to monitor viral spread and prioritize sequences for public health surveillance.

genomicsmachine-learningmethodology
Cancer reports (Hoboken, N.J.) 2026-11-01 Low

Development and Validation of Machine Learning Models Incorporating TGFBIp Expression for Prognosis Prediction in Epithelial Ovarian Cancer.

This study develops and validates machine learning prognostic models for epithelial ovarian cancer that incorporate TGFBIp expression alongside other clinical and molecular features. The authors aim to show that adding TGFBIp improves outcome prediction over existing tools, addressing the heterogeneity of EOC prognosis.

machine-learningcancertranscriptomics
MedComm 2026-11-01 Medium

Metabolic Profiling Combined With Machine Learning for Assessing Clonal Hematopoiesis Status in Chronic Immune Thrombocytopenia: An Exploratory Analysis.

This exploratory study analyzed genomic data from 1,817 chronic immune thrombocytopenia (ITP) patients and found clonal hematopoiesis (CH) in 10.8%, most commonly with DNMT3A mutations. Using untargeted plasma metabolomics in a subset of 539 patients, the authors identified metabolic signatures associated with CH status and applied machine learning to distinguish ITP patients with versus without CH. The work suggests that metabolic profiling combined with machine learning could offer a non-genetic way to assess clonal hematopoiesis in ITP.

metabolomicsmachine-learningclinical
Medicine 2026-10-30 Low

Persistent atrial fibrillation in older adults: Comparison of seven models and development of a gradient boosting machine classifier in a single-center observational study.

This single-center observational study of 572 older adults compared seven classification models for identifying persistent atrial fibrillation and selected a gradient boosting machine classifier as the best performer. The work also characterizes clinical correlates of persistent AF in this population and interprets the chosen model's predictors.

machine-learningclinicalcardiovascular
International journal of molecular sciences 2026-10-28 Medium

Computational Discovery of Plant-Derived Peptides with Therapeutic Potential: A Scoping Review.

This scoping review maps the literature on computational discovery of plant-derived peptides with therapeutic potential, screening over 5000 records and charting 79 sources published between 2021 and 2026. It characterizes the composition of the field—mostly primary studies, with a smaller share of tools and databases—and assesses reporting practices. The review provides a structured overview of methods and gaps in this emerging area.

reviewdrug-discoverymethodology
PloS one 2026-01-01 Medium

Integrating large-scale serum baseline metabolomics and APOE ε4 genotype status for the non-invasive detection of Alzheimer's disease in the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort.

This study tested whether a large panel of serum metabolites, combined with APOE ε4 genotype status, can distinguish Alzheimer's disease patients from cognitively normal individuals in the ADNI cohort. The work evaluates a non-invasive, blood-based approach as an alternative to costly or invasive AD diagnostics. It represents a comparative metabolomics-plus-genetics stratification strategy within a well-characterized cohort.

metabolomicsclinicalmachine-learning
PloS one 2026-01-01 Medium

Machine learning for host-pathogen protein-protein interaction prediction: A tutorial paper.

This tutorial paper presents a step-by-step machine learning pipeline for predicting host-pathogen protein-protein interactions, implemented across six Google Colab notebooks covering dataset preparation, preprocessing, feature extraction, and model training. It aims to make computational HP-PPI prediction accessible to researchers without extensive programming backgrounds by providing a reproducible, hands-on workflow.

machine-learningprotein-structuremethodology
PloS one 2026-01-01 Low

SeqBattNet: A discrete-state physics-informed neural network with aging adaptation for battery modeling.

SeqBattNet is a physics-informed neural network for battery modeling that uses a discrete-state formulation to predict remaining discharge time and energy, addressing limitations of model-based methods (many physical parameters), purely data-driven methods (heavy labeled-data dependence), and prior PINNs (no aging adaptation, continuous state regeneration). The approach is designed to adapt to battery aging while reducing reliance on physical parameters and labeled data.

machine-learningmethodologytool
International journal of general medicine 2026-01-01 Low

Multi-Omics Mendelian Randomization Identifies DDX17 as a Potential Mediator Linking Gut Microbiota-Immune-Metabolism Axis to Alzheimer's Disease.

This study uses multi-omics Mendelian randomization to test whether gut microbiota, immune, and metabolic traits causally influence Alzheimer's disease, and identifies the RNA helicase DDX17 as a potential molecular mediator connecting these peripheral pathways to neurodegeneration. The work nominates a specific host target for follow-up in the microbiota–immune–metabolism axis of AD.

metagenomicsmicrobiomepopulation-genetics
Frontiers in genetics 2026-01-01 Low

Screening of oxidative stress-related signature genes and construction of an interpretable discriminant model for preeclampsia.

This study screens for oxidative stress-related signature genes in preeclampsia and builds an interpretable discriminant model to distinguish PE from normal pregnancies. It also explores how these genes relate to immune regulation in the placenta, aiming to improve molecular understanding and diagnostic potential for this pregnancy complication.

transcriptomicsclinicalmachine-learning
Frontiers in psychiatry 2026-01-01 Medium

A multisource clinical deep learning model for early identification of mental disorders in first-visit psychiatric patients.

This study developed a deep learning model (TUCH-FT) to classify multiple mental disorders in 5,802 first-visit psychiatric patients using demographics, clinical history, and item-level SCL-90 symptom data, with final diagnoses confirmed through longitudinal follow-up and expert evaluation. The model modifies the FT-Transformer architecture to handle heterogeneous clinical data sources separately rather than in a single sequence, aiming to improve early identification when symptoms overlap.

machine-learningclinicaldeep-learning
Frontiers in cellular and infection microbiology 2026-01-01 Medium

An interpretable machine learning model using routine clinical indicators for early pediatric sepsis risk stratification: a multicenter retrospective cohort study.

This multicenter retrospective study developed and validated an interpretable machine learning model that uses routine clinical indicators to stratify pediatric patients for early sepsis risk and to classify Gram-stain pathogen type. The model aims to support earlier recognition of organ dysfunction, aligning with the 2024 Phoenix Sepsis Criteria, and addresses the time lag of blood cultures and the lack of pathogen-level classification in prior models.

machine-learningclinicalmicrobiology
Frontiers in cardiovascular medicine 2026-01-01 Low

Integrative multi-omics and machine learning identify mechanical stimulus-related genes and immune microenvironment remodeling in heart failure.

This study integrates multi-omics data with machine learning to identify mechanical stimulus-related genes (MSRGs) as diagnostic biomarkers for heart failure, and characterizes molecular subtypes of the disease. It further explores how these genes relate to immune microenvironment remodeling and their cellular localization in cardiac tissue. The work proposes MSRG-based biomarkers and immune interactions as potential diagnostic and therapeutic targets in HF.

transcriptomicsmachine-learningclinical
Research and reports in tropical medicine 2026-01-01 Medium

Development of a Machine-Learning Model for Strongyloidiasis Risk Stratification in Migrant Populations.

This study developed and externally validated a machine-learning model to stratify migrants from endemic areas by their risk of strongyloidiasis, a parasitic infection that can cause fatal hyperinfection. The model aims to support targeted screening strategies in migrant populations, where universal screening may be impractical.

machine-learningclinicalmicrobiology
Frontiers in artificial intelligence 2026-01-01 Low

A lightweight multi-branch transformer architecture for Kazakh Sign Language recognition under signer-independent conditions.

The authors introduce a lightweight multi-branch transformer architecture for recognizing isolated Kazakh Sign Language gestures from RGB video, using MediaPipe Holistic to extract 1662-dimensional pose, face, and hand landmark features. They evaluate on a purpose-collected dataset of 39,979 gesture samples across 187 classes from five signers under signer-independent conditions. The work addresses the scarcity of annotated data and benchmarks for low-resource sign languages.

machine-learningdeep-learningmethodology
Frontiers in bioinformatics 2026-01-01 Medium

SelCysML: a hybrid ML-BLAST pipeline for the automated identification and functional annotation of bacterial selenoproteins.

The authors present SelCysML, a hybrid pipeline that combines machine learning with BLAST homology searches to automatically identify bacterial selenoproteins and annotate their functions. The tool addresses widespread misannotation in public databases and the limited interpretability of existing 'black box' predictors, aiming to reduce false positives through integrated functional validation. It targets the challenging problem of detecting selenocysteine incorporation via UGA recoding.

machine-learningproteomicsmethodology
Frontiers in public health 2026-01-01 Low

Dietary amino acid composition and elevated fasting glucose in children and adolescents from four Chinese provinces: a cross-sectional study combining regression and machine learning.

This cross-sectional study examines whether the amino acid composition of children's and adolescents' diets is associated with elevated fasting glucose across four Chinese provinces. The authors combine regression modeling with machine learning and nonlinear dose-response analysis to identify dietary amino acid patterns linked to early markers of impaired glucose metabolism.

metabolomicsmachine-learningclinical
Frontiers in oncology 2026-01-01 Medium

A reproducible open-data framework for interpretable breast cancer diagnosis and cytopathology education.

This study presents a reproducible, open-data framework that uses quantitative nuclear morphology features to classify breast cytopathology as malignant or benign while keeping the diagnostic reasoning interpretable. The framework is designed both as a diagnostic modeling tool and as a teaching resource for medical AI education, linking feature-level interpretation to evidence-based assessment of decisions.

cancermachine-learningclinical

Ask this channel

Answers come only from bioTriage’s indexed literature, scoped to this channel. Always verify before citing.