Machine Learning Algorithms Evaluate Immune Response to Novel Mycobacterium tuberculosis Antigens for Diagnosis of Tuberculosis

Abstract

Rationale: Tuberculosis diagnosis in children remains challenging. Microbiological confirmation of tuberculosis disease is often lacking, and standard immunodiagnostic including the tuberculin skin test and interferon-γ release assay for tuberculosis infection has limited sensitivity. Recent research suggests that inclusion of novel Mycobacterium tuberculosis antigens has the potential to improve standard immunodiagnostic tests for tuberculosis.

Objective: To identify optimal antigen-cytokine combinations using novel Mycobacterium tuberculosis antigens and cytokine read-outs by machine learning algorithms to improve immunodiagnostic assays for tuberculosis.

Methods: A total of 80 children undergoing investigation of tuberculosis were included (15 confirmed tuberculosis disease, five unconfirmed tuberculosis disease, 28 tuberculosis infection and 32 unlikely tuberculosis). Whole blood was stimulated with 10 novel Mycobacterium tuberculosis antigens and a fusion protein of early secretory antigenic target (ESAT)-6 and culture filtrate protein (CFP) 10. Cytokines were measured using xMAP multiplex assays. Machine learning algorithms defined a discriminative classifier with performance measured using area under the receiver operating characteristics.

Measurements and main results: We found the following four antigen-cytokine pairs had a higher weight in the discriminative classifier compared to the standard ESAT-6/CFP-10-induced interferon-γ: Rv2346/47c- and Rv3614/15c-induced interferon-gamma inducible protein-10; Rv2031c-induced granulocyte-macrophage colony-stimulating factor and ESAT-6/CFP-10-induced tumor necrosis factor-α. A combination of the 10 best antigen-cytokine pairs resulted in area under the curve of 0.92 ± 0.04.

Conclusion: We exploited the use of machine learning algorithms as a key tool to evaluate large immunological datasets. This identified several antigen-cytokine pairs with the potential to improve immunodiagnostic tests for tuberculosis in children.

Keywords: cytokines; immune response; interferon-gamma release assay; novel antigens; pediatric tuberculosis.

Figure 1

Whole blood was stimulated with novel antigens and data was analyzed with different machine learning algorithms. (A) Whole blood was stimulated with 11 mycobacterial antigens, left unstimulated and with a positive control, overnight and supernatant was analyzed using Luminex technology to measure 11 different cytokines (B) data was normalized within antigen–cytokine pairs using min-max or mean–std normalization or within patient distribution using the latter only. Data (n = 59) was divided into five equal parts and a classifier discriminating healthy vs. sick children was trained using four subsets and tested on one subset (cross-validation). The algorithm’s parameters were adjusted until performance was optimal. ROC curves were used to measure performance. (C) K-means clustering approach was used to allocate individual data points to three cluster centers randomly. This approach was repeated until optimal data point allocation was reached meaning the sum of the distances from data point to cluster centers is minimized.

Figure 2

Normalization of data contributes to performance of discriminative classifier (A) Cytokine concentrations for individual patients. Results are sorted by patient group and clusters (2, 1 or 0), and antigen–cytokine pairs. Clustering was performed using K-means algorithm. Min–max normalization was applied to cytokine–antigen concentrations, mean–std normalization was applied to between-individual measurements (color change from dark blue to light green represents an increase in relative cytokine concentration). (B) AUROC curve showing the performance of the binary classifier (confirmed/unconfirmed TB and TB infection versus TB exposed) in 59 patients using different normalization methods: min–max and mean–std; normalization of antigen–cytokine pairs; min–max/mean-std combining an antigen–cytokine pair normalization with individual patient normalization.

Figure 3

Effect of normalization of antigen–cytokine pairs and normalization for individual patients (A) Performance of binary classifier using the 10 most important features and applying an antigen–cytokine pair normalization (min–max) and a normalization for individual patients (mean–std) (B) Combination of 10 antigen–cytokine pairs contributing the most to performance of trained discriminative classifier with min–max normalization of antigen–cytokine pairs and mean–std individual patient normalization.