Source Data Underlying Manuscript Figures

<b>Fig1aSupplementaryDataMixture.xlsx</b><br>Source data underlying <b>Figure 1a</b> of the manuscript. In Silico mixtures which are deconvolved.<br>Row names:CellType - Cell type spiked in at a particular fraction<br>SpikePercentage - Percentage cell type is spiked in at<br>TumorContent - Percentage of tumor content added to methylation mixture<br>CancerType - Cancer cell line used as the tumor content<br>Replicate - Replicate number<br><br>Cell type names ending in "_GT" are the ground truth percentages of those cell types<br><br>Rows with "cg" correspond to CpG site on 450k methylation array with the Beta values for each mixture in the columns<br><b>Fig1aSupplementaryDataDeconvolution.xlsx<br></b><br>Source data underlying <b>Figure 1a</b> of the manuscript. Results of deconvolution of in silico mixtures.<br>Column names:Method - Method used for deconvolution<br>CellType - Cell type spiked in at a particular fractionSpikePercentage - Percentage cell type is spiked in at<br>TumorContent - Percentage of tumor content added to methylation mixture<br>CancerType - Cancer cell line used as the tumor content<br>Replicate - Replicate number<br><br>Cell type names ending in "_GT" are the ground truth percentages of those cell types; cell type names not ending in "_GT" are the predicted cell type fractions using the specified method.<br><b>Fig1bSupplementaryDataMixture.xlsx</b><br>Source data underlying <b>Figure 1b</b> of the manuscript. In Vitro mixtures which are deconvolved.<br>Row names:Mixture - In Vitro mixture name -- this corresponds to the cell type fractions in the mixtureTumorContent - Percentage of tumor content added to methylation mixture<br>CancerType - Cancer cell line used as the tumor content<br>NoiseCoefficient - Amount of noise added to the mixture<br>Replicate - Replicate number<br>Cell type names ending in "_GT" are the ground truth percentages of those cell types<br><br>Rows with "cg" correspond to CpG site on 450k methylation array with the Beta values for each mixture in the columns<br><b>Fig1bSupplementaryDataDeconvolution.xlsx</b><br><b><br></b>Source data underlying <b>Figure 1b</b> of the manuscript. Results of deconvolution of in vitro mixtures.<br>Column names:Method - Method used for deconvolution<br>Mixture - In Vitro mixture name -- this corresponds to the cell type fractions in the mixtureTumorContent - Percentage of tumor content added to methylation mixture<br>CancerType - Cancer cell line used as the tumor content<br>NoiseCoefficient - Amount of noise added to the mixture<br>Replicate - Replicate number<br><br>Cell type names ending in "_GT" are the ground truth percentages of those cell types; cell type names not ending in "_GT" are the predicted cell type fractions using the specified method.<br><b>Fig1cSupplementaryDataDeconvolution.xlsx</b><br>Source data underlying <b>Figure 1c</b> of the manuscript. Results of deconvolution of whole blood and engineered mixtures using LTS regression and the new signature matrix.<b><br></b><br>Column names:Mixture - Mixture name -- this corresponds to the cell type fractions in the mixture<br>Cell type names ending in "_GT" are the ground truth percentages of those cell types; cell type names not ending in "_GT" are the predicted cell type fractions using the specified method.<br>RMSE1,RMSE2,R1,R2 -- these correspond to the goodness-of-fit metrics<br><b>Fig2a2bSupplementaryDataDeconvolution.xlsx</b><br>Source data underlying <b>Figures 2a</b> and <b>2b</b> of the manuscript. Results of deconvolution of true positive and true negative samples using LTS regression and the new signature matrix.<br><br>Column names:<br>Sample - Sample GEO accession<br>RMSE1, R1, RMSE2, R2 - goodness of fit metrics which are plotted in Figures 2a and 2bCell type names - the predicted cell type fractions using LTS and the new signature<br>Tissue - the annotated tissue (or tissue of origin)<br><b>Fig2c2dSupplementaryDataMixture.txt</b><br>Source data underlying <b>Figures 2c</b> and <b>2d</b> of the manuscript. Mixtures generated from pairwise combinations of true positive and true negative samples with varying fractions of true negative content.<br>Rows with "cg" correspond to CpG site on 450k methylation array with the Beta values for each mixture in the columns.<br><br>Columns are the mixture name with the amount of true negative.<br><b>Fig2c2dSupplementaryDataDeconvolution.xlsx</b><br>Source data underlying <b>Figures 2c</b> and <b>2d</b> of the manuscript. Results of deconvolution of mixtures generated from pairwise combinations of true positive and true negative samples with varying fractions of true negative content using LTS regression and the new signature matrix.<br><br>Column names:<br>Mixture - The mixture name with the amount of true negative fraction in the mixture indicated in the name<br>RMSE1, R1, RMSE2, R2 - goodness of fit metrics which are plotted in Figures 2c and 2dCell type names - the predicted cell type fractions using LTS and the new signature<br><b>Fig2eSupplementaryDataCorrelations.xlsx</b><br>Source data underlying <b>Figure 2e</b> of the manuscript. Results of deconvolution of TCGA samples which were used to find correlations between cell types fractions or goodness-of-fit metrics and the consensus purity estimate (CPE).<br>Column names:<br>Sample - TCGA sample IDRMSE1, R1, RMSE2, R2 - goodness of fit metricsCell type names - the predicted cell type fractions using LTS and the new signatureCPE - consensus purity estimate for the TCGA sample<br><b>Fig3SupplementaryDataTraining.xlsx</b><br>Source data underlying <b>Figure 3</b> of the manuscript. TCGA samples used to train and test the random forest model used to predict tumor purity.<b><br></b><br>Column names:<br>Sample - TCGA sample ID<br>CancerType - TCGA cancer type<br>ESTIMATE, ABSOLUTE, LUMP, IHC, CPE - different methods for estimating tumor purityRMSE1, R1, RMSE2, R2 - goodness of fit metricsCell type names - the predicted cell type fractions using LTS and the new signature<br><b>Fig3SupplementaryDataEvaluation.xlsx</b><br>Source data underlying <b>Figure 3</b> of the manuscript. TCGA samples used to evaluate the random forest model used to predict tumor purity. These samples were completely held out from the training and testing of the model.<b><br></b><br>Column names:<br>Sample - TCGA sample ID<br>CancerType - TCGA cancer type<br>ESTIMATE, ABSOLUTE, LUMP, IHC, CPE - different methods for estimating tumor purityRMSE1, R1, RMSE2, R2 - goodness of fit metricsCell type names - the predicted cell type fractions using LTS and the new signature<br><b>Fig5bSupplementaryDataCoxPH.xlsx</b><br>Source data underlying <b>Figure 5b</b> of the manuscript. Cox PH was run on a pan-cancer deconvolution of TCGA to get hazard ratios for each cell type in each cancer.<b><br></b><br>Column names:<br>Cancer - TCGA cancer typeCellType - Cell type considered in the Cox PHFDR - adjusted p-values for significance of the cell typeexp.coef - Hazard ratio