A new bioinformatic pipeline allows the design of small, targeted gene panels for efficient TMB estimation

Abstract

Background: The tumor mutation burden (TMB) is emerging as a prognostic and predictive marker for the response to immune checkpoint blockade (ICB) drugs. We aimed to develop a new method for the definition of gene panels that can precisely estimate the TMB with a considerably lower amount of genome.

Methods: We developed a bioinformatic pipeline which allows the design of gene panels suited for TMB estimation. The method is particularly efficient in optimizing the balance between the precision of the TMB estimate and the length of the gene panel created. We tested in silico the efficiency of different panels obtained with our method in an independent cohort of patients with lung adenocarcinoma (LUAD). We also compared in the same cohort of patients the performance of our panels with the performance of existing gene panels.

Results: We designed a 0.080 Megabases (Mb) long gene panel which estimated TMB in an independent LUAD cohort with an acceptable precision (adjusted R2=0.745; Spearman ρ = 0.827, Pearson ρ = 0.864). The panel showed 0.89 accuracy in the identification of TMB-high patients (25/28 patients, CI: 0.73 – 0.96). Every unitary increase of our TMB estimate was associated with lower risk of disease progression (univariate analysis: HR = 0.78; CI: 0.65-0.93; p = 0.006; multivariate analysis: HR = 0.8; CI: 0.63-1.01; p = 0.0621). Different existing panels of less than 1 Mb long showed a lower adjusted R2 when compared to our gene panel (Table). Two other commercial panels of 1.9 Mb and 1.1 Mb showed a similar adjusted R2 to panels of the same lengths built with our method; nevertheless, they showed a lower accuracy in TMB-high patients definition (ROC curves AUC of 0.862, 0.870, 0.946 and 0.967 were observed, respectively, for the commercial 1.9 Mb panel, the commercial 1.1 Mb panel, our 0.080 Mb panel and our 2.0 Mb panel).