• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Thursday, November 19, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

Machine Learning Algorithms Accurately Calibrate Multiplex Serology for Helicobacter Pylori Diagnosis

by
November 12, 2026
in Health
Reading Time: 6 mins read
0
Machine Learning Algorithms Accurately Calibrate Multiplex Serology for Helicobacter Pylori Diagnosis

Machine Learning Algorithms Accurately Calibrate Multiplex Serology for Helicobacter Pylori Diagnosis

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

Researchers evaluating serological diagnostic methodologies within the China Kadoorie Biobank have established that supervised classification algorithms can successfully calibrate continuous multiplex serology assays against established immunoblot standards for identifying Helicobacter pylori infection. By applying five distinct machine learning models including Bayesian additive regression trees, multidimensional monotone Bayesian additive regression trees, random forests, elastic net regularization, and logistic regression, the investigation demonstrated exceptional discriminative accuracy. Every evaluated algorithm achieved an area under the receiver operating characteristic curve of at least 95 percent when predicting serostatus against binary immunoblot results. Furthermore, predictive performance remained remarkably consistent between cross-validated subcohort training data and an independent validation cohort of gastric cancer cases. The study resolves a persistent methodological obstacle in large-scale biobank research by providing a mathematically robust framework to translate continuous multidimensional antigen antibody readouts into calibrated posterior probabilities of bacterial exposure without discarding vital diagnostic variation.

Accurate determination of infection status is critical because Helicobacter pylori represents the primary risk factor for gastric malignancies, accounting for an estimated 89 percent of non-cardia gastric cancer occurrences worldwide and roughly 810,000 annual cancer cases globally. The bacterium colonizes the stomach, where chronic colonization causes gastric cancer and mucosa-associated lymphoid tissue lymphoma while showing potential associations with colon cancer. In epidemiological studies, bacterial presence is typically assessed with serologic assays measuring immunoglobulin G antibodies. However, advanced atrophic gastritis, a precursor of gastric cancer, frequently causes antibody level reductions prior to clinical diagnosis. This attenuation creates false-negative serological classifications among cancer cases, diminishing apparent statistical associations in retrospective studies. Prospective investigations like the China Kadoorie Biobank, which recruited half a million Chinese adults aged 30 to 79 years between 2004 and 2008, circumvent this diagnostic bias by collecting biological samples many years prior to disease onset.

Within biobank settings, biological sample volume and laboratory throughput dictate operational choices between assay platforms. The reference Helicoblot 2.1 immunoblot assay evaluates antibodies to seven Helicobacter pylori antigens, including cytotoxin-associated gene A, vacuolating cytotoxin A, and urease alpha subunit, yielding binary positive or negative serostatus. Past evaluations confirm that Helicoblot 2.1 correctly classified 98 percent of infected patients and 86 percent of uninfected controls against gastric biopsy polymerase chain reaction, 100 percent of active infections against histology, and demonstrated 98.9 percent sensitivity against stool testing. Crucially, immunoblot identified 92 percent of past infections, indicating it captures cumulative lifetime exposure rather than discriminating active from resolved colonization. However, immunoblot protocols require substantial plasma volumes and intensive labor, rendering them impractical for whole-biobank testing. In contrast, high-throughput multiplex serology simultaneously quantifies antibody responses across 12 bacterial antigens using fluorescence-coded beads on Luminex 200 analyzers, consuming minimal plasma while yielding continuous median fluorescence intensities.

Translating continuous fluorescence measurements into reliable serostatus classifications previously relied on rigid cut-offs that introduced notable statistical deficiencies. Prior studies defined threshold boundaries for each antigen, classifying participants as infected only if four or more antigens exceeded their assigned values. Methodologists have repeatedly criticized dichotomizing continuous biological data because collapsing continuous measurements discards substantial predictive information. Furthermore, requiring four arbitrary positive antigen tests treats all biological targets equally, ignoring differences in their evidentiary strength. In reality, individual antigens provide unequal diagnostic insight regarding true infection status. Treating multiplex calibration as a supervised learning challenge resolves these limitations. By modeling continuous measurements directly, classification algorithms estimate antigen influence weights and generate continuous probability outputs rather than rigid binary decisions. These calibrated infection probabilities circumvent information loss, reflect disparate antigen evidentiary contributions, and provide flexible outputs where investigators can later select pragmatic decision thresholds suited to specific clinical or epidemiological research aims.

The calibration analysis utilized data from an established case-cohort investigation within the China Kadoorie Biobank. Cohort exclusions removed participants with missing blood specimens, cancer diagnoses prior to baseline recruitment, cancer diagnoses occurring within two years of follow-up, death or loss to follow-up within two years, and inconsistent records. From the remaining eligible cohort, researchers selected a random subcohort of 2,000 individuals stratified by age and sex. A subsample of 498 subcohort participants underwent testing with both immunoblot and multiplex serology to construct the model training dataset. The calibrated algorithms were subsequently validated on 922 gastric cancer cases from the eligible cohort who similarly received both assays. Due to the causal link between Helicobacter pylori and gastric malignancy, seroprevalence was 75.7 percent in the subcohort training group and reached 93.5 percent among cancer cases. Before model calibration, all twelve continuous antigen measurements were log-transformed to control for observed variance and standardized to mean zero and variance one.

The researchers evaluated five distinct supervised classification algorithms spanning tree ensembles and regression models. Decision trees recursively partition data based on whether antigen predictors fall above or below specific thresholds, naturally generalizing earlier classification criteria. Because individual decision trees exhibit high variance and instability, the authors implemented tree ensemble methods: random forest, Bayesian additive regression trees, and multidimensional monotone Bayesian additive regression trees. Tree ensembles possess the notable advantage of automatically accounting for complex multi-antigen interactions without requiring manual model specification. To benchmark these non-parametric methods, the authors also fitted standard logistic regression and regularized elastic net models. While unregularized logistic regression can become unstable when modeling intricate interaction terms, elastic net incorporates shrinkage penalties to retain stability. Models were trained using tenfold cross-validation on subcohort data, where nested fivefold cross-validation tuned hyperparameters. Hyperparameter selection minimized root-mean-square error for Bayesian additive regression trees, minimized misclassification error for elastic net, and maximized classification accuracy for random forest and monotone trees.

Algorithmic performance was evaluated using three complementary metrics: the Brier score, the logarithmic score, and the area under the receiver operating characteristic curve. The Brier score, calculating the mean squared prediction error between estimated probabilities and true binary serostatus, served as the primary performance benchmark. Because it is a strictly proper scoring rule bounded between zero and one, the Brier score rewards probability distributions that closely match observed infection states, with lower values denoting superior predictive calibration. The logarithmic score, or cross-entropy loss, measures Kullback-Leibler divergence between predicted probabilities and observed binary outcomes across non-negative values. Like the Brier score, it is negatively oriented but imposes severe penalties on overconfident erroneous predictions. Finally, the area under the receiver operating characteristic curve assessed discriminative capacity, quantifying the probability that the classifier assigns a higher seropositive likelihood to a truly infected individual than to an uninfected individual across varying decision thresholds.

All five calibrated models achieved high discriminative capacity, with area under the curve estimates exceeding 95 percent on both training and test cohorts. On the subcohort training dataset, Bayesian additive regression trees attained the best overall performance according to Brier score and area under the curve, whereas elastic net achieved the lowest logarithmic score. Ranked by decreasing Brier performance on training data, the models ordered as Bayesian additive regression trees, elastic net, random forest, logistic regression, and monotone trees. Evaluated on the independent testing set of 922 gastric cancer cases, Bayesian additive regression trees maintained superior predictive calibration, recording the lowest Brier score at 0.0745 and lowest logarithmic score at 0.2313. Elastic net achieved the highest area under the curve at 0.9716, while monotone trees achieved 97 percent discriminative accuracy. Differences across algorithms remained modest, confirming that modern machine learning effectively maps continuous multiplex serology to reference immunoblot serostatus.

Analyzing antigen importance revealed substantial concordance across distinct statistical models, clarifying the relative evidentiary contributions of individual bacterial targets. Importance measures were derived from splitting inclusion proportions across 100 Bayesian additive regression tree replications with 20 trees each, mean accuracy decreases across 500 random forest trees, and standardized logistic regression coefficients. The three algorithms exhibited 67 percent ranking concordance across the 12 antigens, displaying perfect agreement on the four most influential targets: cytotoxin-associated gene A, conserved hypothetical secreted protein, chaperonin GroEL, and outer membrane protein HP1564. Cytotoxin-associated gene A emerged universally as the single most influential antigen, consistent with its known high immunogenicity, persistence after eradication, and established association with gastric malignancy risk. The models also agreed on hydantoin utilization protein and cinnamyl alcohol dehydrogenase among the least influential predictors. Conversely, although vacuolating cytotoxin A and urease alpha subunit are measured by both assays, neither antigen ranked among the top predictors.

Several methodological considerations and epidemiological boundaries qualify the interpretation and broader generalizability of these findings. Small performance score differences among algorithms highlight the need for future statistical hypothesis testing to establish whether performance advantages reflect genuine algorithmic differences rather than random sampling variation. Furthermore, study generalizability depends directly on specific epidemiological characteristics of the China Kadoorie Biobank, where control participants exhibited high infection seroprevalence of 75.7 percent and infections predominantly involved cytotoxin-associated gene A-positive strains. In low-prevalence populations or settings where bacterial strains lack cytotoxin-associated gene A, diagnostic trade-offs between sensitivity and specificity will differ, requiring dedicated recalibration on local population subsamples. The current models also focused exclusively on serological biomarkers, omitting demographic variables that fall outside standard laboratory procedures. In conclusion, the study demonstrates that supervised machine learning algorithms, particularly Bayesian additive regression trees, provide a principled, effective methodology for calibrating high-throughput multiplex serology against immunoblot benchmarks in large prospective cohorts.

Subject of Research: Supervised machine learning calibration of multiplex serological assays for Helicobacter pylori infection

Article Title: Calibrating multiplex serology for Helicobacter pylori

Article References: Dankwa, E. A., Plummer, M., Chapman, D., Jeske, R., Butt, J., Hill, M., Waterboer, T., Millwood, I. Y., Yang, L., & Kartsonaki, C. (2025). Calibrating multiplex serology for Helicobacter pylori. Diagnostic and Prognostic Research, 9(1), Article 17. https://doi.org/10.1186/s41512-025-00202-x

Image Credits: AI Generated

DOI: 10.1186/s41512-025-00202-x

Keywords: Helicobacter pylori, multiplex serology, immunoblot assay, China Kadoorie Biobank, Bayesian additive regression trees, random forest, elastic net, logistic regression, multidimensional monotone BART, Brier score, area under the curve, gastric cancer risk, CagA antigen, seroprevalence calibration, Helicoblot 2.1, HcpC antigen, GroEL chaperonin, Luminex bead assay

News Source: Ophelia Keating. (November 12, 2026). Machine Learning Algorithms Accurately Calibrate Multiplex Serology for Helicobacter Pylori Diagnosis. Scienmag.

Tags: area under the curveBayesian Additive Regression TreesBrier scoreCagA antigenChina Kadoorie Biobankelastic netgastric cancer riskGroEL chaperoninHcpC antigenHelicobacter pyloriHelicoblot 2.1immunoblot assaylogistic regressionLuminex bead assaymultidimensional monotone BARTmultiplex serologyRandom Forestseroprevalence calibration
Share12Tweet7Share2ShareShareShare1

Related Posts

Mixed and Dedifferentiated Endometrial Carcinomas Share Clonal Roots and Hypermutated Backgrounds

Mixed and Dedifferentiated Endometrial Carcinomas Share Clonal Roots and Hypermutated Backgrounds

November 19, 2026
Isometric Handgrip Exercise Fails to Lower Blood Pressure in Trained Hypertensive Women

Isometric Handgrip Exercise Fails to Lower Blood Pressure in Trained Hypertensive Women

November 19, 2026

Three Cell Death Biomarkers May Help Identify Ischemic Stroke in Blood

November 19, 2026

Depth Perception Often Remains Impaired After Central Serous Chorioretinopathy Therapy

November 19, 2026

POPULAR NEWS

  • Corncrake Flagship Status Does Not Guarantee Local Recognition in Rural Romania

    Corncrake Flagship Status Does Not Guarantee Local Recognition in Rural Romania

    29 shares
    Share 12 Tweet 7
  • Pre-Rollout Survey Finds Eighteen Percent HPV Vaccine Uptake Among Chinese College Women

    29 shares
    Share 12 Tweet 7
  • Military Physicians Issue Fewer Medical Exemptions as Decision Fatigue Mounts Across Workdays

    29 shares
    Share 12 Tweet 7
  • Counseling and Mobile App Increase Postpartum Modern Contraceptive Uptake in Jordan

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Mixed and Dedifferentiated Endometrial Carcinomas Share Clonal Roots and Hypermutated Backgrounds

Knowledge Transfer and Blockchain Capability in Saudi Supply Chain Financing

Isometric Handgrip Exercise Fails to Lower Blood Pressure in Trained Hypertensive Women

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 88 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.