Skip to contents

Applying a DLBCLone model to new data

Preamble

This guide will demonstrate the steps for training a custom DLBCLone model and applying it to classify new samples.

Preparing your test data

To use a trained DLBCLone model on your data, you need (at a minimum) either a MAF file or a mutation status matrix for the genes (features) that were used to generate the model. Here, you will learn how to generate a mutation status matrix using one of our published cohorts, which was sequenced with the LySeqST panel. If you have translocation status for BCL2, BCL6 etc and want to include that in the model, you will also need a second table that includes the translocation status for each sample. We generally refer to this as the metadata table. Together, your MAF and metadata files will be used to assemble the feature matrix that will be used for UMAP and KNN approximation on your samples.

SV status

For this example, we will extract the BCL2 and BCL6 status directly from the output of LymphGen. We will also use the LymphGen classification to evaluate our accuracy. Importantly, you do not need a LymphGen result from your test data but it can be useful to sanity check your UMAP and your model performance. The MAF and metadata for these samples (in the form of a LymphGen output) are both bundled with this package. The code below loads these files and restructures a few columns to make them compatible with downstream functions.

library(GAMBLR.open)
library(GAMBLR.utils) 
library(tidyverse)
library(GAMBLR.predict)

lymphgen_test = "extdata/lyseq_validation/LySeqST_only_unaugmented_vaf.lymphgen_calls.no_cnvs.with_sv.no_A53.tsv"
valid_metadata <- readr::read_tsv(
  system.file(lymphgen_test,package = "GAMBLR.predict")) %>%
  rename(
    sample_id = Sample.Name,
    lymphgen = Subtype.Prediction
  ) %>%
  mutate(
    BCL2.Translocation = if_else(BCL2.Translocation == "yes", "POS", BCL2.Translocation),
    BCL2.Translocation = if_else(BCL2.Translocation == "no", "NEG", BCL2.Translocation),
    BCL2.Translocation = if_else(BCL2.Translocation == "Not Available", "NA", BCL2.Translocation),
    BCL6.Translocation = if_else(BCL6.Translocation == "yes", "POS", BCL6.Translocation),
    BCL6.Translocation = if_else(BCL6.Translocation == "no", "NEG", BCL6.Translocation),
    BCL6.Translocation = if_else(BCL6.Translocation == "Not Available", "NA", BCL6.Translocation)
  )
maf_test = "extdata/lyseq_validation/LySeqST_merged_unaugmented_vaf.maf"
valid_maf_vaf_dir <- readr::read_tsv(system.file(maf_test,package = "GAMBLR.predict")) %>%  
  mutate(
    sample_id = Tumor_Sample_Barcode
  )
#convert to S3 object for compatability
valid_maf_vaf <- create_maf_data(
    maf_df = valid_maf_vaf_dir,
    genome_build = "grch37"
  ) %>%
    mutate(Chromosome = as.character(Chromosome))

Synonymous mutations

If the model you’re using used base-3 encoding (0,1,2), that means synonymous mutations are considered for some genes. You will have to know up-front (or find out) which genes this applies to, so you can encode your data consistently. If your MAF file doesn’t contain any synonyous mutations, be sure to use a model that was trained only on coding mutations.

metadata 

knitr::kable(head(valid_metadata))
sample_id Copy.Number BCL2.Translocation BCL6.Translocation Model.Used Confidence.BN2 Confidence.EZB Confidence.MCD Confidence.N1 Confidence.ST2 BN2.Feature.Count EZB.Feature.Count MCD.Feature.Count N1.Feature.Count ST2.Feature.Count BN2.Features EZB.Features MCD.Features N1.Features ST2.Features lymphgen
00-14595_tumorA Not Available POS POS NOCGH 0.79 0.99 0.00 0.07 0.03 3 3 1 0 2 BCL6FusionCatcherORInHouse_FUSION,DTX1_MUTATION,BTG2_SubSyn BCL2TransFISHMUTFusion_CompUp,EZH2_SubMUTATION,SOCS1_MUTATION BTG2_SubSyn NA SOCS1_MUTATION,ITPKB_MUTATION EZB
00-14595_tumorB Not Available POS POS NOCGH 0.79 0.99 0.00 0.07 0.68 3 3 3 0 3 BCL6FusionCatcherORInHouse_FUSION,DTX1_MUTATION,BTG2_SubSyn BCL2TransFISHMUTFusion_CompUp,EZH2_SubMUTATION,SOCS1_MUTATION TBL1XR1_MUTATION,PIM2_SubSyn,BTG2_SubSyn NA SGK1_TRUNC,SOCS1_MUTATION,ITPKB_MUTATION EZB
00-14595_tumorC Not Available POS POS NOCGH 0.79 0.99 0.00 0.07 0.68 3 4 3 0 3 BCL6FusionCatcherORInHouse_FUSION,DTX1_MUTATION,BTG2_SubSyn BCL2TransFISHMUTFusion_CompUp,EZH2_SubMUTATION,KMT2D_TRUNC,SOCS1_MUTATION TBL1XR1_MUTATION,PIM2_SubSyn,BTG2_SubSyn NA SGK1_TRUNC,SOCS1_MUTATION,ITPKB_MUTATION EZB
00-14595_tumorD Not Available POS NEG NOCGH 0.17 0.98 0.00 0.07 0.00 2 2 2 0 0 DTX1_MUTATION,BTG2_SubSyn BCL2TransFISHMUTFusion_CompUp,EZH2_SubMUTATION PIM2_SubSyn,BTG2_SubSyn NA NA EZB
00-15201_tumorA Not Available NEG NEG NOCGH 0.07 0.05 0.05 0.07 0.15 1 1 3 0 3 TNFAIP3_TRUNC KMT2D_TRUNC PIM1_Synon,BTG1_SubSyn,HLA-A_MUTATION NA CD83_Synon,DUSP2_MUTATION,WEE1_Synon Other
00-15201_tumorB Not Available NEG NEG NOCGH 0.18 0.11 0.01 0.99 0.00 2 1 3 1 0 TNFAIP3_TRUNC,KLF2_Synon EP300_MUTATION HLA-B_MUTATION,PIM2_SubSyn,HLA-A_MUTATION NOTCH1_MUTATION NA N1

lyseq features

The LySeqST sequencing panel includes the following genes. In order to train a classifier compatible with this panel, we need to subset the training data to ensure only these genes are included in training.

lyseq_genes <- c(
  "TNFRSF14", "SPEN", "ID3", "ARID1A", "RRAGC",
  "BCL10", "CD58", "NOTCH2", "FCGR2B", "FCRLA",
  "FAS", "CCND1", "BIRC3", "ATM", "KMT2D",
  "STAT6", "BTG1", "FOXO1", "B2M", "MAP2K1",
  "IDH2", "CHD2", "CREBBP", "SOCS1", "IL4R",
  "PLCG2", "TP53", "STAT5B", "STAT3", "CD79B",
  "GNA13", "BCL2", "TCF3", "S1PR2", "JAK3",
  "MEF2B", "DNMT3A", "XPO1", "CXCR4", "SF3B1",
  "PTPN1", "XBP1", "EP300", "MYD88", "SETD2",
  "RHOA", "NFKBIZ", "TBL1XR1", "KLHL6", "TET2",
  "PIM1", "CCND3", "TMEM30A", "PRDM1", "SGK1",
  "TNFAIP3", "CARD11", "POT1", "BRAF", "EZH2",
  "UBR5", "MYC", "NOTCH1", "TRAF2", "P2RY8",
  "BTK", "TP73", "NOL9", "SEMA4A", "PRRC2C",
  "BTG2", "ITPKB", "SEC24C", "EDRF1", "WEE1",
  "MPEG1", "ETS1", "DTX1", "SETD1B", "NFKBIA",
  "ZFP36L1", "CIITA", "MBTPS1", "IRF8", "ACTG1",
  "KLHL14", "MED16", "CD70", "JUNB", "KLF2",
  "CD79A", "DYSF", "DUSP2", "BCL2L1", "PRDM15",
  "RFTN1", "OSBPL10", "EIF4A2", "BCL6", "IRF4",
  "FOXC1", "CD83", 
  "HLA-A", "HLA-B", "PRRC2A", "TBCC", "INTS1",
  "ACTB", "PIK3CG", "PRKDC", "TOX", "CDKN2A",
  "GRHPR", "DDX3X", "PIM2", "UBE2A", "ETV6",
  "MS4A1", "CD19", "HNRNPD", "NFKBIE", "TMSB4X"
)

Making a compatible feature matrix

You create a mutation feature matrix with assemble_genetic_features. This can include mutation status and structural variant status for BCL2, BCL6, and MYC. By default, synonymous mutations will be encoded as 1 and SV and coding mutations will be encoded as 2. The genes for which synonymous mutations are considered are specified separately. The genes we commonly use are shown in the example below.

ashm_genes = c("ACTB","ACTG1","BCL2","BCL6","BIRC3",
     "BTG1","BTG2","CD83","CIITA","CXCR4","DDX3X",    
     "DTX1","DUSP2","ETS1","ETV6","GRHPR","ID3", 
     "IL4R","IRF4","IRF8","ITPKB","KLF2","KLHL6",
     "MEF2B","MS4A1","MYC","NFKBIZ","OSBPL10","PIM1",
     "PIM2","PTPN1","RFTN1","S1PR2","SGK1","SOCS1",
     "TMSB4X","WEE1","XBP1","ZFP36L1")

lyseq_valid = assemble_genetic_features(
  these_samples_metadata = valid_metadata,
  sv_from_metadata = c(BCL2 = 'BCL2.Translocation', BCL6 = 'BCL6.Translocation'),
  genes=lyseq_genes,
  maf_with_synon = valid_maf_vaf,
  hotspot_genes = "MYD88",
  synon_genes=ashm_genes
)

Load the training data 

all_full_status_train <- readr::read_tsv(system.file("extdata/all_full_status.tsv",package = "GAMBLR.predict")) %>%
  tibble::column_to_rownames("sample_id")

dlbcl_meta <- readr::read_tsv(system.file("extdata/dlbcl_meta_with_dlbclass.tsv",package = "GAMBLR.predict"))

dlbcl_meta_clean <- filter(dlbcl_meta,
                          lymphgen %in% c("MCD","EZB","BN2","N1","ST2","Other"))

Before training, the mutation matrix from the bundled training data must be subset to exclude any features that are not part of your panel.

ncol(all_full_status_train)
[1] 202
ncol(lyseq_valid)
[1] 125
lyseq_train <- all_full_status_train %>%
  select(all_of(names(lyseq_valid)))

Create training data UMAP 

mu_lyseq_train = make_and_annotate_umap(
  lyseq_train,
  dlbcl_meta_clean,
  core_features = list(
    ST2=c("SGK1","DUSP2","TET2","SOCS1"),
    N1="NOTCH1",
    EZB=c("EZH2","BCL2_SV"),
    MCD=c("MYD88HOTSPOT","CD79B","PIM1"),
    BN2=c("BCL6_SV","NOTCH2","SPEN","CD70")
  )
)

Visualize UMAP

Visually confirm the separation of the training samples based on their LymphGen class labels.

make_umap_scatterplot(mu_lyseq_train$df)

Train the model

Train and optimize our model using the training data with our custom feature set.

lyseq_features_optimized = DLBCLone_optimize_params(  
  mu_lyseq_train$features, 
  dlbcl_meta_clean,
  umap_out = mu_lyseq_train,
  truth_classes = c("MCD","EZB","BN2","N1","ST2","Other"),
  min_k=11,
  max_k=21
)

lyseq_features_optimized <- DLBCLone_activate( # Activates optimized model setting the UMAP projection
  lyseq_features_optimized, 
  force = TRUE
)
lyseq_features_optimized <- DLBCLone_activate( # Activates optimized model setting the UMAP projection
  lyseq_features_optimized, 
  force = TRUE
)
transforming each data point individually using the provided UMAP model. This will take some time.
done
model given to function

Training accuracy

Check the classification accuracy of the optimized classifier (DLBCLone_wo) on the training data set.

make_alluvial(lyseq_features_optimized,
  pred_column="DLBCLone_wo")
Warning: Duplicated aesthetics after name standardisation: fill
Warning in to_lodes_form(data = data, axes = axis_ind, discern =
params$discern): Some strata appear at multiple axes.
Warning in to_lodes_form(data = data, axes = axis_ind, discern =
params$discern): Some strata appear at multiple axes.

Sanity check feature incidence in the test data 

pred_check <- DLBCLone_predict(
  mutation_status = lyseq_valid,
  optimized_model = lyseq_features_optimized,
  check_frequencies = TRUE
)
pred_check$plot

Apply the model to the test data 

pred_valid_lyseq_all <- DLBCLone_predict(
  mutation_status = lyseq_valid,
  optimized_model = lyseq_features_optimized
)

knitr::kable(head(pred_valid_lyseq_all$prediction))
sample_id predicted_label confidence other_score neighbor_id neighbor distance label other_neighbor vote_labels weighted_votes neighbors_other neighborhood_otherness other_weighted_votes total_w pred_w V1 V2 .id MCD_NN_count EZB_NN_count BN2_NN_count N1_NN_count ST2_NN_count Other_NN_count top_group MCD_score EZB_score BN2_score N1_score ST2_score Other_score top_score_group top_group_score top_group_count Other_count by_vote by_vote_opt by_score score_ratio by_score_opt DLBCLone_w DLBCLone_wo
00-14595_tumorA EZB 1.0000000 0.0000000 00-14595_tumorC,17-33848_tumorB,06-30025T,DLBCL11428T,09-41114T,03-11110T,06-30145T,SP116620,DLBCL-MC_F349_BJJ-Tumor,16-13732T,DLBCL11572T,CAR_154_PreCART,SU-DHL-4,05-25439T,DLBCL11656T,08-13706T,CAR_178_PreCART,05-25674T,06-18449_tumorB 1,354,121,1106,172,47,122,1331,626,321,1218,508,2425,88,1240,140,515,89,109 0.128,0.142,0.145,0.149,0.152,0.156,0.195,0.205,0.212,0.223,0.227,0.244,0.251,0.251,0.251,0.26,0.271,0.274,0.288 EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB EZB 95.7664551143418 0 0.0000000 0.000000 95.76646 95.76646 -0.8899906 4.3290272 1 0 19 0 0 0 0 EZB 0.000000 95.76646 0.00000 0.00000 0.000000 0.000000 EZB 95.76646 19 0 19 EZB EZB Inf EZB EZB EZB
00-14595_tumorB EZB 1.0000000 0.0000000 09-41114T,03-11110T,00-14595_tumorC,17-33848_tumorB,DLBCL11428T,DLBCL-MC_F349_BJJ-Tumor,06-30025T,DLBCL11572T,SU-DHL-4,08-13706T,05-25439T,CAR_178_PreCART,CAR_154_PreCART,DLBCL11656T,SP116620,06-30145T,CAR_177_PreCART,06-18449_tumorB,DLBCL11567T 172,47,1,354,1106,626,121,1218,2425,140,88,515,508,1240,1331,122,514,109,1213 0.113,0.115,0.13,0.141,0.158,0.163,0.181,0.188,0.194,0.207,0.21,0.215,0.219,0.229,0.229,0.247,0.259,0.26,0.263 EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB EZB 103.880530854753 0 0.0000000 0.000000 103.88053 103.88053 -0.8807368 4.2731309 2 0 19 0 0 0 0 EZB 0.000000 103.88053 0.00000 0.00000 0.000000 0.000000 EZB 103.88053 19 0 19 EZB EZB Inf EZB EZB EZB
00-14595_tumorC EZB 1.0000000 0.0000000 08-13706T,CAR_177_PreCART,CLC03454,SU-DHL-4,DLBCL11567T,DLBCL11572T,DLBCL-MC_F349_BJJ-Tumor,SP116659,CLC00259,CAR_154_PreCART,06-18449_tumorB,DLBCL-MC_F300_TBF-Tumor,DLBCL10459T,CAR_178_PreCART,02-13135T,05-32762T,FL1002T2,09-41114T,DLBCL10465T 139,513,356,2424,1212,1217,625,1338,188,507,108,621,831,514,28,93,10,171,836 0.132,0.159,0.161,0.163,0.165,0.167,0.172,0.18,0.181,0.192,0.193,0.197,0.197,0.197,0.228,0.229,0.231,0.233,0.235 EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB EZB 102.42382208477 0 0.0000000 0.000000 102.42382 102.42382 -0.9343416 4.1505504 3 0 19 0 0 0 0 EZB 0.000000 102.42382 0.00000 0.00000 0.000000 0.000000 EZB 102.42382 19 0 19 EZB EZB Inf EZB EZB EZB
00-14595_tumorD EZB 1.0000000 0.0409382 190806-PL01-PL02,DLBCL11666T,DLBCL-MC_F358_WDS-Tumor,DLBCL11573T,05-25674T,10-15025T,Aph_13_PreCART,DLBCL11558T,FL1010T2,SU-DHL-6,SP116620,06-15922T,SP194216,CAR_276_PreCART,DLBCL11575T,SP193957,171290-PL01,170867-PL02,15-11280_CLC02281 450,1249,627,1219,89,177,469,1206,449,2427,1331,105,1301,526,1220,1307,461,460,285 0.118,0.126,0.154,0.157,0.159,0.166,0.173,0.18,0.202,0.204,0.207,0.21,0.21,0.234,0.236,0.252,0.253,0.253,0.255 EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB,EZB FL1003T2 EZB 101.424039965503 1 0.0526316 4.152118 101.42404 101.42404 -0.6999627 4.4204650 4 0 19 0 0 0 0 EZB 0.000000 101.42404 0.00000 0.00000 0.000000 4.152118 EZB 101.42404 19 1 19 EZB EZB 24.4270604 EZB EZB EZB
00-15201_tumorA BN2 0.4536479 8.6321139 Reddy_3551T,Reddy_2620T,Reddy_2789T,Reddy_3904T,DLBCL11518T,SP116690,DLBCL-c_D_1158-Tumor,Reddy_2650T,1011-2020_FFPE,DLBCL-RICOVER_978-Tumor,Reddy_2603T,Reddy_2534T,Reddy_3616T,QC2-30T,DLBCL10856T,DLBCL10919T,Reddy_705T,Reddy_3849T,H041_T_WGS 2076,1768,1862,2315,1184,1343,813,1788,218,761,1756,1714,2112,1503,941,999,2411,2279,600 0.189,0.233,0.484,0.491,0.569,0.575,0.624,0.679,0.681,0.683,0.722,0.739,0.821,0.857,0.859,0.865,0.888,0.91,0.952 BN2,BN2,ST2,ST2,BN2,ST2,ST2,EZB,EZB,BN2,EZB,BN2,BN2,EZB,EZB,EZB,EZB,EZB,MCD 1102503091T,Reddy_3739T,Reddy_2147T,1102503011T,Reddy_3865T,DFCIDL008-Tumor,DLBCL10977T,1011-2018_FFPE,Reddy_3843T,Reddy_3819T,1011-2018_plasma,Reddy_2634T,DLBCL10933T,DLBCL-c_D_1120-Tumor,00-22011_tumorB,DLBCL11002T,DLBCL11595T,Reddy_2206T,1001-2030_FFPE,Reddy_3845T,1102503013T,Reddy_3733T,DLBCL10955T,OZM073-017_Diagnosis,Reddy_3988T,Reddy_3914T,Reddy_3519T,Reddy_3487T,Reddy_2996T,Reddy_2138T,03-10440_tumorB,03-10440_tumorA,DLBCL11468T,DLBCL-RICOVER_314-Tumor,Reddy_3580T,DLBCL11255T,DLBCL10455T,DLBCL10943T,Reddy_2463T,SP116627,Reddy_3431T,Reddy_2253T,Reddy_702T,Reddy_2123T,Reddy_3943T,Reddy_3556T,Reddy_3412T,Reddy_2714T,Reddy_2479T,Reddy_2719T,Reddy_3981T,02-15630_tumorB,Reddy_3978T,86-30788_tumorA,Reddy_813T,Reddy_2157T,Reddy_2889T,Seq_1_PreCART,94-15772_tumorA,Reddy_2928T,Reddy_2680T,QC2-5T,Reddy_2194T,DLBCL10499T,98-22532T,DLBCL10908T,DLBCL-c_D_1152-Tumor,Reddy_3470T,Reddy_3946T,94-15772_tumorB,DLBCL11180T,DLBCL11681T,Reddy_3654T,Reddy_2688T,1001-2030_plasma BN2,EZB,MCD,ST2 15.3910313542603,10.0403628209901,1.05047600249142,7.44539197428758 75 3.9473684 132.857136 33.92726 15.39103 -1.2391076 0.3538165 5 1 8 6 0 4 0 EZB 1.050476 10.04036 15.39103 0.00000 7.445392 132.857136 BN2 15.39103 8 75 8 Other BN2 0.1158465 Other BN2 Other
00-15201_tumorB N1 0.9709068 0.0466497 DLBCL10973T,DLBCL11648T,10-39294_tumorA,DLBCL10520T,DLBCL11005T,14-33798_tumorB,DLBCL11503T,DLBCL11513T,10-39294_tumorB,DLBCL10844T,CLC00246,DLBCL-MAYO_DLBCL_234-Tumor,1102502995T,Reddy_3982T,DLBCL10915T,c_D_1050_T,DLBCL10952T,Reddy_3997T,DLBCL11439T 1048,1235,186,885,1078,276,1172,1180,187,929,315,1442,1418,2384,995,1433,1028,2399,1115 0.173,0.183,0.224,0.264,0.284,0.288,0.289,0.293,0.389,0.412,0.478,0.486,0.5,0.528,0.529,0.542,0.571,0.613,0.623 N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,N1,ST2 1102503029T N1,ST2 53.5304488302436,1.6040359180294 1 0.0526316 2.497182 55.13448 53.53045 0.7030904 -4.2771802 6 0 0 0 18 1 0 N1 0.000000 0.00000 0.00000 53.53045 1.604036 2.497182 N1 53.53045 18 1 18 N1 N1 21.4363427 N1 N1 N1

Sanity check the UMAP

If the UMAP is working as it should be, you should see a similar level of separation of your test samples in UMAP space. Since we have LymphGen classifications for our test samples, we can check this. In order to visualize these, the metadata will need to be attached to the UMAP projection of the new samples. What you should hopefully see is roughly the same placement of your samples on this scatterplot relative to the training samples.

projected_with_meta = left_join(pred_valid_lyseq_all$projection,
select(valid_metadata,sample_id,lymphgen))
Joining with `by = join_by(sample_id)`
head(projected_with_meta)
        sample_id         V1         V2 lymphgen
1 00-14595_tumorA -0.8899906  4.3290272      EZB
2 00-14595_tumorB -0.8807368  4.2731309      EZB
3 00-14595_tumorC -0.9343416  4.1505504      EZB
4 00-14595_tumorD -0.6999627  4.4204650      EZB
5 00-15201_tumorA -1.2391076  0.3538165    Other
6 00-15201_tumorB  0.7030904 -4.2771802       N1
unique(projected_with_meta$lymphgen)
 [1] "EZB"         "Other"       "N1"          "BN2/N1"      "BN2"
 [6] "MCD"         "ST2"         "EZB/MCD/ST2" "EZB/ST2"     "BN2/EZB"
[11] "BN2/MCD"     "EZB/MCD"    
#drop composites
projected_with_meta = filter(projected_with_meta,!grepl("/",lymphgen))
make_umap_scatterplot(projected_with_meta)

make_neighborhood_plot 

test_sample = "92-38267_tumorB"

make_neighborhood_plot(
  single_sample_prediction_output = pred_valid_lyseq_all,
  this_sample_id = test_sample,
  prediction_in_title = TRUE,
  add_circle = TRUE,
  label_column = "DLBCLone_wo"
)

nearest_neighbor_heatmap 

nearest_neighbor_heatmap(
  this_sample_id = test_sample,
  DLBCLone_model = pred_valid_lyseq_all,
  font_size = 8
)

Accuracy Assessment

Finally, lets assess the accuracy of the validation set by comparing dlbclone_predict predictions with the LymphGen results that were previously generated.

pred_vs_original = left_join(pred_valid_lyseq_all$prediction,valid_metadata)

#collapse composites into group-COMP
pred_vs_original = tidy_lymphgen(pred_vs_original,lymphgen_column_in = "lymphgen",lymphgen_column_out = "lymphgen_tidy")

#convert composites into their primary class
pred_vs_original = mutate(pred_vs_original,lymphgen_basic = gsub("-COMP","",lymphgen_tidy))

make_alluvial(list(predictions=pred_vs_original),
  pred_column = "DLBCLone_wo",
  group_order = c("MCD","BN2","ST2","EZB","N1","Other"),
  truth_column = "lymphgen_basic")

Oncoplot

Another way to visualize the results is via an Oncoplot (using GAMBLR.viz::prettyOncoplot).

library(GAMBLR.viz)
ashm_genes = intersect(ashm_genes,lyseq_genes)

silent_show = rep("Silent",length(ashm_genes))
names(silent_show)=ashm_genes
# add hotspot annotation
valid_maf_vaf$hot_spot = NA
valid_maf_vaf = mutate(valid_maf_vaf,hot_spot=ifelse(Hugo_Symbol=="MYD88" & HGVSp_Short=="p.L265P",TRUE,FALSE))
prettyOncoplot(these_samples_metadata = pred_vs_original, 
  maf_df = valid_maf_vaf,
  metadataColumns = c("lymphgen_basic","DLBCLone_w","DLBCLone_wo"),
  minMutationPercent = 1,
  cluster_rows = T,
  highlightHotspots = T,
  genes = lyseq_genes,
  pctFontSize = 3,
  fontSizeGene = 3,
  hideSideBarplot = T,
  simplify_annotation = T,
  hide_annotations = c("DLBCLone_wo","DLBCLone_w"),
  include_noncoding = as.list(silent_show),
  sortByColumns = c("DLBCLone_wo","lymphgen_basic"))