Incorporation of cut-point robustness in the discovery and evaluation of molecular cancer biomarkers using SCRIBE (Survival Cut-point Reporting and Integrative Biomarker Evaluation)

Precision oncology relies on the use of biomarkers for matching cancer patients with specific therapies aiming to maximize efficacy and minimize toxicity. Although a large number of candidate biomarkers are reported, only few of them show adequate performance in independent studies and are implemented clinically. Major limitations of molecular biomarker studies in oncology include the utilization of individual cohorts and the selection of single, and arbitrary cut-points to stratify patients that increase the risk of false discovery and data overfitting. To address this challenge, we introduce a novel concept for biomarker discovery, development, prioritization and validation incorporating the cut-point robustness along ranges of consecutive cut-points as a critical parameter. We also report SCRIBE, a novel and user-friendly automated computational tool for simultaneous analysis, and discovery of biomarkers and their impact along cut-point ranges in independent datasets. To illustrate the performance of our approach, we used SCRIBE to assess the biomarker potential of the mRNA expression levels of 13 HLA class II antigen-presentation machinery genes in baseline tumor samples from three independent cohorts of patients with lung adenocarcinoma. Our analysis identified HLA-DQB1 as the HLA class-II gene with the most robust prognostic association. The analysis of two cohorts identified the optimal cut-point range between the 44 th and 57 th percentiles. To define a single cut-point, we selected the percentile with the strongest hazard ratio effect, which corresponded to the 49 th percentile. We applied this cut-point to the three cohorts and found that higher levels of HLA-DQB1 were significantly associated with longer overall survival. Using Cox proportional hazard models including HLA-DQB1 and stage, both features were significantly and independently associated with survival in the cohorts. Finally, we performed an unsupervised discovery analysis of the whole transcriptome incorporating the cut-point robustness and identified genes with prominent and previously unreported association with prognosis in patients with lung adenocarcinoma. Notably, AVEN emerged as the top gene with a dramatic negative survival effect. We provide a novel conceptual framework for biomarker studies incorporating the cut-point robustness as a key property. We also provide a novel integrative computational tool for automated analysis of multiple biomarkers incorporating relevant cut-point ranges and other properties to enhance the impact of biomarker studies. We envision SCRIBE as a valuable tool for researchers and clinicians seeking to improve the reproducibility and applicability of biomarker studies and to support the development and clinical use of existing and novel biologically driven anti-cancer therapies.