Loading...
Download data
Welcome to DecoNFlow Benchmarking
A benchmark of DNA methylation deconvolution methods for tumoral fraction estimation using DecoNFlow
bioRxiv 2025 Nov 27. doi: 10.1101/2025.11.27.688590
Benchmarking plots: WGBS-TT
Table of Contents
Boxplots of the predictions for each tumoral fraction
The plot shows predicted tumoral fractions as boxplots for each deconvolution tool (grouped by DMR tool) at a selected expected TF (marked by a red dashed line), with tools ranked left-to-right by their average absolute deviation from the expected TF, so that more accurate tools appear earlier.
Performance (NRMSE)
The plot shows NRMSE values for a selected deconvolution tool across expected tumoral fractions (X-axis) and DMR tools (color), allowing comparison of prediction error normalized by expected value, with lower points indicating better performance.
Heatmap of expected tumoral fraction vs deconvolution tools
The plot shows log2-scaled NRMSE values (tile color) for each deconvolution tool across expected tumoral fractions, with tools ranked left-to-right by their NRMSE at expected TF = 0.0001, where lower values indicate better performance.
AUC-ROC at different tumoral fractions
The plot shows ROC curves and AUC values for each selected deconvolution tool (faceted), across multiple low tumoral fractions (0.0001 to 0.05), where each line color represents a fraction and higher curves indicate better classification performance.
This interactive plot shows ROC curves and AUC values for a selected deconvolution tool across multiple low tumoral fractions (0.0001 to 0.5). Each curve represents a different fraction, and the AUC value is indicated at FPR = 0 for each. Hover over lines and points to view detailed sensitivity, specificity, and AUC metrics. Higher AUC values and curves closer to the top-left indicate better classification performance.
Tools RMSE
The plot shows RMSE values for each deconvolution tool (Y-axis) across selected DMR tools (colored points) at a specific expected tumoral fraction, with tools sorted top-to-bottom by their mean RMSE, so that lower (better) RMSE tools appear at the top.
Limit of detection (LoD)
The plot shows predicted tumoral fractions across increasing expected fractions for a selected tool and DMR method, with boxplots and Wilcoxon test significance markers comparing each level to 0 to assess the lowest fraction at which signal becomes statistically distinguishable from noise.
Final ranking of the tools
To get a unique aggregated metric per combination, individual metrics (RMSE, AUC, SCC, and LoD) are min–max scaled across deconvolution tools at each combination analysed and aggregated into a final score, and then min–max scaled again across tools: \(\small \text{Score} = \text{RMSE} + \text{AUC} + \text{SCC} + \text{LoD}\)
Benchmarking plots: RRBS-TT
Table of Contents
Boxplots of predictions for each tumoral fraction
The plot shows predicted tumoral fractions as boxplots for each deconvolution tool (grouped by DMR tool), at a selected expected TF (marked by a red dashed line), allowing comparison of prediction accuracy and variability across methods.
Performance (NRMSE)
The plot shows NRMSE values for a selected deconvolution tool across expected tumoral fractions (X-axis) and DMR tools (color), allowing comparison of prediction error normalized by expected value, with lower points indicating better performance.
Heatmap of expected tumoral fraction vs deconvolution tools
The plot shows log2-scaled NRMSE values (tile color) for each deconvolution tool across expected tumoral fractions, with tools ranked left-to-right by their NRMSE at expected TF = 0.0001, where lower values indicate better performance.
AUC-ROC for different tumoral fractions
The plot shows ROC curves and AUC values for each selected deconvolution tool (faceted), across multiple low tumoral fractions (0.0001 to 0.05), where each line color represents a fraction and higher curves indicate better classification performance.
This interactive plot shows ROC curves and AUC values for a selected deconvolution tool across multiple low tumoral fractions (0.0001 to 0.5). Each curve represents a different fraction, and the AUC value is indicated at FPR = 0 for each. Hover over lines and points to view detailed sensitivity, specificity, and AUC metrics. Higher AUC values and curves closer to the top-left indicate better classification performance.
Tools RMSE
The plot shows RMSE values for each deconvolution tool (Y-axis) across selected DMR tools (colored points) at a specific expected tumoral fraction, with tools sorted top-to-bottom by their mean RMSE, so that lower (better) RMSE tools appear at the top.
Limit of detection (LoD)
The plot shows predicted tumoral fractions across increasing expected fractions for a selected tool and DMR method, with boxplots and Wilcoxon test significance markers comparing each level to 0 to assess the lowest fraction at which signal becomes statistically distinguishable from noise.
Final ranking of deconvolution tools
To get a unique aggregated metric per combination, individual metrics (RMSE, AUC, SCC, and LoD) are min–max scaled across deconvolution tools at each combination analysed and aggregated into a final score, and then min–max scaled again across tools: \(\small \text{Score} = \text{RMSE} + \text{AUC} + \text{SCC} + \text{LoD}\)
Benchmarking plots: RRBS-CL
Table of Contents
Boxplot of predictions for each tumor fraction
The plot shows predicted tumoral fractions as boxplots for each deconvolution tool (grouped by DMR tool), at a selected expected TF (marked by a red dashed line), allowing comparison of prediction accuracy and variability across methods.
Performance (NRMSE)
The plot shows NRMSE values for a selected deconvolution tool across expected tumoral fractions (X-axis) and DMR tools (color), allowing comparison of prediction error normalized by expected value, with lower points indicating better performance.
Heatmap of expected tumoral fraction vs deconvolution tools
The plot shows log2-scaled NRMSE values (tile color) for each deconvolution tool across expected tumoral fractions, with tools ranked left-to-right by their NRMSE at expected TF = 0.0001, where lower values indicate better performance.
AUC-ROC for different tumoral fractions
The plot shows ROC curves and AUC values for each selected deconvolution tool (faceted), across multiple low tumoral fractions (0.0001 to 0.05), where each line color represents a fraction and higher curves indicate better classification performance.
This interactive plot shows ROC curves and AUC values for a selected deconvolution tool across multiple low tumoral fractions (0.0001 to 0.5). Each curve represents a different fraction, and the AUC value is indicated at FPR = 0 for each. Hover over lines and points to view detailed sensitivity, specificity, and AUC metrics. Higher AUC values and curves closer to the top-left indicate better classification performance.
Tools RMSE
The plot shows RMSE values for each deconvolution tool (Y-axis) across selected DMR tools (colored points) at a specific expected tumoral fraction, with tools sorted top-to-bottom by their mean RMSE, so that lower (better) RMSE tools appear at the top.
Limit of detection (LoD)
The plot shows predicted tumoral fractions across increasing expected fumoral fractions for a selected tool and DMR method, with boxplots and Wilcoxon test significance markers comparing each level to 0 to assess the lowest fraction at which signal becomes statistically distinguishable from noise.
Final ranking of deconvolution tools
To get a unique aggregated metric per combination, individual metrics (RMSE, AUC, SCC, and LoD) are min–max scaled across deconvolution tools at each combination analysed and aggregated into a final score, and then min–max scaled again across tools: \(\small \text{Score} = \text{RMSE} + \text{AUC} + \text{SCC} + \text{LoD}\)
Deconvolution tools
DMR tools
Contact Us
Contributors
Shiny App
Sofie Van de Velde
R Code
Edoardo Giuili
Supervision
Prof. Dr. Ir. Katleen De Preter
Prof. Dr. Celine Everaert
