Objective:To construct a signature for identifying active tuberculosis (TB) based on the relative expression orderings (REOs) of gene expression within a single sample.Methods:Using peripheral whole blood samples from 75 active TB and 69 latently infected individuals from four datasets as the training set, and highly stable REO patterns were extracted from the gene expression profile of the two groups of samples. Then, the gene pairs that reversed the REO pattern between the two groups were selected, and each gene pair was ranked in descending order based on their reversal degree. Finally, the top k gene pairs with the highest classification accuracy were selected as the signature for independent dataset validation. Results:A signature composed of seven gene pairs, denoted as 7-GPS, was constructed from the training set. The accuracy rate for 7-GPS to distinguish active TB from latently infected samples was 88.89%, and the accuracy rate for distinguishing active TB from normal samples was 90.09%. In the mixed validation data from different detection platforms, the AUC value for distinguishing active TB from latently infected samples was 0.914 (95% CI: 0.881-0.948), and the AUC value for distinguishing active TB from normal samples was 0.934 (95% CI: 0.904-0.964). In addition, the four genes ETV7, BATF2, ANKRD22 and CARD17P from this signature tended to be highly expressed in peripheral blood samples of active TB, and their expression values were significantly related to the duration of anti-tuberculosis treatment in clinical. Conclusion:The 7-GPS signature is robust and suitable for individualized analysis of a single peripheral blood sample. It has certain clinical application potential.

Objective:To construct a signature for identifying active tuberculosis (TB) based on the relative expression orderings (REOs) of gene expression within a single sample.Methods:Using peripheral whole blood samples from 75 active TB and 69 latently infected individuals from four datasets as the training set, and highly stable REO patterns were extracted from the gene expression profile of the two groups of samples. Then, the gene pairs that reversed the REO pattern between the two groups were selected, and each gene pair was ranked in descending order based on their reversal degree. Finally, the top k gene pairs with the highest classification accuracy were selected as the signature for independent dataset validation. Results:A signature composed of seven gene pairs, denoted as 7-GPS, was constructed from the training set. The accuracy rate for 7-GPS to distinguish active TB from latently infected samples was 88.89%, and the accuracy rate for distinguishing active TB from normal samples was 90.09%. In the mixed validation data from different detection platforms, the AUC value for distinguishing active TB from latently infected samples was 0.914 (95% CI: 0.881-0.948), and the AUC value for distinguishing active TB from normal samples was 0.934 (95% CI: 0.904-0.964). In addition, the four genes ETV7, BATF2, ANKRD22 and CARD17P from this signature tended to be highly expressed in peripheral blood samples of active TB, and their expression values were significantly related to the duration of anti-tuberculosis treatment in clinical. Conclusion:The 7-GPS signature is robust and suitable for individualized analysis of a single peripheral blood sample. It has certain clinical application potential.