Fig. 2: DELFI-TF accurately predicted cfDNA tumor burden in patients with metastatic colorectal cancer.

a Model schematic. Plasma aliquots were collected from patients and used for cfDNA isolation. From each time point of patients with tissue-confirmed RAS/BRAF mutations, duplicate cfDNA samples were utilized for ddPCR and low-coverage WGS. WGS fragment-sequencing statistics were calculated per sample at a given time point. A random forest model was trained against the MAFs called by ddPCR readouts of the tumor-specific RAS/BRAF variants in all longitudinal cfDNA samples to generate the DELFI-TF values. b Patients with mCRC exhibit a wide range of DELFI-TF values at baseline (T0; n = 128) and reduced tumor fractions at the first time point after treatment commencement (T1; n = 151). Tumor fractions increase from treatment commencement (T1) to progression (n = 73). Non-cancer controls exhibit remarkably low DELFI-TF values (n = 153). No significant difference was observed between mutant type (MT) and wild-type (WT) samples for either time point (T0, p = 6.10e-01, T1, p = 5.14e-01, progression, p = 4.57e-01, two-sided Wilcoxon rank-sum). c DELFI-TF strongly correlates with detectable MAF values measured by ddPCR (n = 177, r = 0.90, p = 8.83e-65, two-sided Pearson correlation). Plasma time points with undetectable MAF (n = 132; n = 5 at T0, 40 at T1, 28 at T2, 59 at T3–T9) are indicated in red and included those with high DELFI-TF values in patients which were independently validated as having measurable tumor burden. The middle hinge in the boxplots corresponds to the median, while the lower and upper hinges correspond to the first and third quartiles. The upper whisker extends from the hinge to the largest value no further than the 1.5 * interquartile range from the hinge. Ribbons around the regression line in correlation plots represent the 95% confidence level interval for predictions.