Figure 3.
Fragmentation landscapes of cell-free DNA reveal lymphoma-related disparities and provide a mutation-independent relapse predictor in end-of-therapy samples. (A) Histograms demonstrating differences in fragment length distribution between mutated (red) and wild-type (WT, blue) fragments in pretreatment plasma samples from all the patients with single nucleotide variants detected and pretreatment cfDNA sample available. (B) Heatmap showing fragment density according to fragment length (y-axis) in all pretreatment cfDNA profiles (x-axis). Examples of different patterns are demonstrated on the right with smoothened histograms. I: Case#1 with prominent mononucleosomal pattern vs Case#2 with the most prominent dinucleosomal pattern. II: Case#3 histogram favoring longer fragment lengths (perinucleosomal) vs Case#4 histogram favoring shorter fragment lengths (subnucleosomal). III: Case#5 with bifurcation of the mononucleosomal fragments that reflect the difference between mutated and WT fragments (histogram below; red, mutated fragments; blue, reference sequence fragments). (C) Scatterplot of the first 2 dimensions of the principal component analysis (PCA) of fragment profiles with all the cfDNA samples in the study. Colors of the dots represent sample groups indicated in the legend below. (D) Box plots demonstrate the differences in the first 3 components of the PCA between cured and relapsing patients in the end-of-therapy plasma samples. (E) Line graph showing correlation of the first 3 components with fragment size histogram. Below: 9 histogram bin size windows selected based on their correlation with fragment size and mutually exclusive pattern with each other. (F) Heatmaps demonstrate the concordance between the 2 independent predictors of recurrence among the relapsing and nonrelapsing patients. Monte Carlo, mutation-MRD-based relapse predictor with -log10 transformed P values; Random forest, classifier predictions for relapse likelihood 0% to 100%. (G) Receiver operating characteristics curves for the phasing-aware mutation-based (‘Monte Carlo’) and fragmentation-based (‘Random forest’) predictors for recurrence. AUC, area under curve. (H) Kaplan-Meier survival estimates for FFS according to the combined mutation-based MRD test and fragmentation-based random forest prediction results in the end-of-therapy plasma samples.

Fragmentation landscapes of cell-free DNA reveal lymphoma-related disparities and provide a mutation-independent relapse predictor in end-of-therapy samples. (A) Histograms demonstrating differences in fragment length distribution between mutated (red) and wild-type (WT, blue) fragments in pretreatment plasma samples from all the patients with single nucleotide variants detected and pretreatment cfDNA sample available. (B) Heatmap showing fragment density according to fragment length (y-axis) in all pretreatment cfDNA profiles (x-axis). Examples of different patterns are demonstrated on the right with smoothened histograms. I: Case#1 with prominent mononucleosomal pattern vs Case#2 with the most prominent dinucleosomal pattern. II: Case#3 histogram favoring longer fragment lengths (perinucleosomal) vs Case#4 histogram favoring shorter fragment lengths (subnucleosomal). III: Case#5 with bifurcation of the mononucleosomal fragments that reflect the difference between mutated and WT fragments (histogram below; red, mutated fragments; blue, reference sequence fragments). (C) Scatterplot of the first 2 dimensions of the principal component analysis (PCA) of fragment profiles with all the cfDNA samples in the study. Colors of the dots represent sample groups indicated in the legend below. (D) Box plots demonstrate the differences in the first 3 components of the PCA between cured and relapsing patients in the end-of-therapy plasma samples. (E) Line graph showing correlation of the first 3 components with fragment size histogram. Below: 9 histogram bin size windows selected based on their correlation with fragment size and mutually exclusive pattern with each other. (F) Heatmaps demonstrate the concordance between the 2 independent predictors of recurrence among the relapsing and nonrelapsing patients. Monte Carlo, mutation-MRD-based relapse predictor with -log10 transformed P values; Random forest, classifier predictions for relapse likelihood 0% to 100%. (G) Receiver operating characteristics curves for the phasing-aware mutation-based (‘Monte Carlo’) and fragmentation-based (‘Random forest’) predictors for recurrence. AUC, area under curve. (H) Kaplan-Meier survival estimates for FFS according to the combined mutation-based MRD test and fragmentation-based random forest prediction results in the end-of-therapy plasma samples.

Close Modal

or Create an Account

Close Modal
Close Modal