Figure 1.

Experimental workflow. Illustration of all experiments and methodology with annotations. Step 1 involves the discovery of novel candidate biomarkers for discerning actionable versus non-actionable nodules utilizing HuProt™ microarrays with our “Discovery” cohort (n = 62 total samples; divided into 10 sample pools). A total of 26 candidate biomarkers were identified: 20 from microarray studies and 6 from our previous work. In Step 2, custom Luminex immunobead assays are developed for the candidate biomarkers identified in Step 1 and used to assess their performance against our Biomarker Development Cohort (n = 841). Each marker was statistically evaluated for its individual value for discerning actionable versus non-actionable nodules. In Step 3, the data from the Biomarker Development Cohort (n = 841) is split into three cohorts: Training, Validation 1, and Validation 2. The random forest algorithm was used to develop a biomarker panel based on the optimal combination of six features. The model’s performance characteristics for discerning actionable versus non-actionable cases were evaluated and optimized using Validation cohorts 1 and 2.