Deciphering a TB-related DNA methylation biomarker and constructing a TB diagnostic classifier

Mengyuan Lyu; Jian Zhou; Lin Jiao; Yili Wang; Yanbing Zhou; Hongli Lai; Wei Xu; Binwu Ying

doi:10.1016/j.omtn.2021.11.014

Deciphering a TB-related DNA methylation biomarker and constructing a TB diagnostic classifier

Mol Ther Nucleic Acids. 2021 Nov 19:27:37-49. doi: 10.1016/j.omtn.2021.11.014. eCollection 2022 Mar 8.

Authors

Mengyuan Lyu^{1

2}, Jian Zhou^{2

3}, Lin Jiao^{1

2}, Yili Wang^{1

2}, Yanbing Zhou^{1

2}, Hongli Lai², Wei Xu^{4

5}, Binwu Ying^{1

2}

Affiliations

¹ Department of Laboratory Medicine, West China Hospital, Sichuan University, No. 37, Guoxue Alley, Chengdu, Sichuan 610041, China.
² West China School of Medicine, Sichuan University, Chengdu, Sichuan 610041, China.
³ Department of Thoracic Surgery, West China Hospital, Sichuan University, Chengdu, Sichuan 610041, China.
⁴ Department of Biostatistics, Princess Margaret Cancer Centre, University Health Network, 10-511, 610 University Avenue, Toronto, ON M5G 2M9 Canada.
⁵ Dalla Lana School of Public Health, University of Toronto, Toronto, ON M5T 3M7 Canada.

Abstract

We systemically identified tuberculosis (TB)-related DNA methylation biomarkers and further constructed classifiers for TB diagnosis. TB-related DNA methylation datasets were searched through October 3, 2020. Limma and DMRcate were employed to identify differentially methylated probes (DMPs) and regions (DMRs). Machine learning methods were used to construct classifiers. The performance of the classifiers was evaluated in discovery datasets and a prospective independent cohort. Eighty-nine DMPs and 24 DMRs were identified based on 67 TB patients and 45 healthy controls from 4 datasets. Nine and three DMRs were selected by elastic net regression and logistic regression, respectively. Among the selected DMRs, two regions (chr3: 195635643-195636243 and chr6: 29691631-29692475) were differentially methylated in the independent cohort (p = 4.19 × 10^-5 and 0.024, respectively). Among the ten classifiers, the 3-DMR logistic regression classifier exhibited the strongest performance. The sensitivity, specificity, and area under the curve were, respectively, 79.1%, 84.4%, and 0.888 in the discovery datasets and 64.5%, 90.3%, and 0.838 in the independent cohort. The differential diagnostic ability of this classifier was also assessed. Collectively, these data showed that DNA methylation might be a promising TB diagnostic biomarker. The 3-DMR logistic regression classifier is a potential clinical tool for TB diagnosis, and further validation is needed.

Keywords: DNA methylation; biomarker; classifier; molecular diagnosis; tuberculosis.