Fig. 1

Overview of the study. This study used 12 somatic fresh SEQC2 T/N (Tumor/Normal) biological replicates sequenced at six centers. Three centers have three replicates, and three have one replicate. We developed three replicate-based consensus approaches (within-center, cross-center, and all centers) using the results of three replicate centers. In the within-center approach, we used replicates from the same center; in the cross-center approach, we used replicates from different centers; and in the all centers approach, we used the results of all replicates as input. We accepted the declared high-confidence variants as ground truth and extracted the precision, recall, and F1 scores of these approaches. In the second part of the study, we trained machine learning models based on the results of these approaches (detected somatic variants). We used the information on the first five chromosomes in the training set and the remaining chromosomes in the test set. We also trained machine learning models based on declared high-confidence variants (instead of developed approaches). Finally, we extracted the precision, recall, and F1 score of these trained machine learning models by accepting declared high-confidence somatic variants as ground truth