Tumor fraction correlations at TFBSs

For 191 MBC ULP samples with >0.1 tumor fraction, nucleosome profiling with and without GC correction was performed on the top 10,000 sites for each of 377 transcription factors (TFs) using nucleosome sized fragments (100–200 bp). For each TF, the relationship between central coverage and tumor fraction was modeled using scipy.stats.linregress⁸⁴ producing a Pearson correlation (r) and line of best fit (scipy version 1.7.1). Pearson p-values for each feature type were adjusted using a Benjamini-Hochberg FDR correction. Root mean squared error (RMSE) was calculated from the line of best fit. This was performed both before and after GC correction as illustrated for LYL1 in Fig. 2e. For all 377 TFs, the RMSE values before and after GC correction were compared using a Wilcoxon signed-rank test (two-sided). This same procedure was applied to test the benefit of an additional mappability correction step and an additional copy number correction step.