Enhancing HLA-B27 antigen detection: Leveraging machine learning algorithms for flow cytometric analysis

Area of Science:

• Immunogenetics
• Computational Biology
• Clinical Diagnostics

Background:

• Human leukocyte antigen B27 (HLA-B27) is strongly associated with spondylarthropathies.
• Flow cytometry is commonly used for HLA-B27 detection but can yield equivocal results due to cross-reactivity.
• A "gray zone" in flow cytometry results necessitates further classification for accurate HLA-B27 diagnosis.

Purpose of the Study:

• To apply machine learning (ML) methods for classifying equivocal HLA-B27 antigen expression data.
• To evaluate the performance of various ML algorithms in distinguishing positive from negative HLA-B27 results in borderline cases.
• To enhance the accuracy and reliability of HLA-B27 testing in clinical settings.

Main Methods:

• Selected 99 equivocal samples for HLA-B27 antigen testing analysis.
• Utilized flow cytometry and polymerase chain reaction for sample analysis.
• Trained and validated logistic regression (LR), decision tree (DT), random forest (RF), and light gradient boost method (GBM) using flow cytometry histogram features.

Main Results:

• All evaluated ML algorithms demonstrated high accuracy, sensitivity, specificity, and predictive values.
• The random forest algorithm showed the best performance with an AUC of 0.92 on the tested sample set.
• AUC values for light GBM, DT, and LR were 0.88, 0.89, and 0.89, respectively, with no significant statistical differences between algorithms.
• Gradient boost methods may be less effective on smaller sample sizes.

Conclusions:

• Machine learning algorithms can effectively classify equivocal HLA-B27 data, reducing uncertain results.
• Implementation of ML in HLA-B27 testing can minimize false negatives and false positives, particularly where genetic testing is unavailable.
• ML offers a valuable tool for improving the diagnostic precision of HLA-B27 testing in clinical laboratories.