Bug fixing and writing tests for release version 1.0

DESCRIPTION

@@ -1,10 +1,10 @@
 Package: BarrettsProgressionRisk
-Title: Risk progression from Barrett's Esophagus to Adenocarcinoma
+Title: Prognositc adenocarcinoma risk prediction for Barrett's Esophagus 
 Type: Package
-Version: 0.5.0.1
+Version: 1.0
 Date: 2020-01-10
 Authors@R: person("Sarah","Killcoyne", email="skillcoy@ebi.ac.uk", role=c("aut","cre"))
-Description: Process and predict risk of progression from shallow WGS data. 
+Description: Process and predict risk of adenocarcinoma progression from shallow whole-genome sequencing data. 
 License: GPL-3
 Encoding: UTF-8
 LazyData: true

R/plot_utilities.R

@@ -86,6 +86,9 @@ plotCorrectedCoverage<-function(brr, as=c('plot','list')) {
 showPredictionCalibration<-function(df=NULL) {
   if (is.null(df)) df = BarrettsProgressionRisk:::be_model$pred.confidence
 
+  if (!is.factor(df$Risk))
+    df = df %>% mutate(Risk = factor(Risk, levels=names(riskColors()), ordered=T))
+
   mm = range(df[c('r1','r2')])
 
   cuts = seq(mm[1], mm[2], by=df$r1[2])
@@ -97,7 +100,7 @@ showPredictionCalibration<-function(df=NULL) {
 
   ggplot(df, aes(mn, perc)) + 
     geom_rect(aes(xmin=r1, xmax=r2, ymin=0,ymax=1, fill=Risk), alpha=0.6) + 
-    scale_fill_manual(values=riskColors(), limits=levels(df$Risk) ) +
+    scale_fill_manual(values=unlist(riskColors()), limits=levels(df$Risk) ) +
     geom_vline(xintercept=cuts[2:(length(cuts)-1)], color='grey88') +
     geom_smooth(method='lm',formula=y~x, color='grey39', linetype='dashed', fill='grey88', size=0.5, fullrange=T) + 
     geom_point() + geom_errorbar(aes(ymin=ci.low, ymax=ci.high), size=0.5, width=0.01) +
@@ -163,17 +166,22 @@ patientRiskTilesPlot<-function(brr, col='Endoscopy', direction=c('fwd','rev')) {
   }
 
   gej.dist = grep('Distance',colnames(preds),value=T)
-
   if (length(gej.dist) > 0 & !is.factor(preds[[gej.dist]])) {
     preds[[gej.dist]] = fct_rev(factor(preds[[gej.dist]], ordered=T))
-  } else if (length(gej.dist <= 0)) {
+  } else if (length(gej.dist) <= 0) {
     preds[[gej.dist]] = 1
   }
-  preds = preds %>% mutate_if(is.numeric, list(~factor(.,ordered=T)))
+
+  if (!is.factor(preds[[col]])) {
+    preds[[col]] = factor(preds[[col]], ordered=T)
+  }
+
+  preds = preds %>% mutate_if(is.numeric, list(~factor(.,ordered=T))) %>%
+    mutate(Risk = factor(Risk, levels=names(riskColors()), ordered=T))
 
   p = ggplot(preds, aes_(as.name(col), as.name(gej.dist))) +
     geom_tile(aes(fill=Risk), color='white',size=2) + 
-    scale_fill_manual(values=riskColors(), limits=names(riskColors())) +
+    scale_fill_manual(values=unlist(riskColors()), limits=names(riskColors())) +
     labs(y='Esophageal Location (GEJ...)')
 
   if (dir == 'rev') p = p + scale_x_discrete(limits=rev(levels(preds[[col]])))
@@ -200,14 +208,23 @@ patientEndoscopyPlot<-function(brr) {
   if (length(which(class(brr) %in% c('BarrettsRiskRx'))) <= 0)
     stop("BarrettsRiskRx required")
 
+  printRisk <- function(x,low,high, risk='Unknown') {
+    img = switch(risk, 
+                 'High'='img/Human_body_silhouette-RED.png',
+                 'Moderate'='img/Human_body_silhouette-YELLOW.png',
+                 'Low'='img/Human_body_silhouette-BLUE.png',
+                 'Unknown'='img/Human_body_silhouette-GREY.png')
+    paste0(paste0(c("",rep(paste0('<img src="',img,'" alt="%" width="8"></img>'), x)), collapse=""), ' <b>',x, '%</b> (',low,'%-',high,'%)')
+  } 
+
   preds = absoluteRiskCI(brr)
   preds = preds %>% rowwise() %>% dplyr::mutate( img=printRisk(Probability*100,CI.low*100,CI.high*100,Risk) )
 
   ggplot(preds, aes(Endoscopy, Probability)) + ylim(0,1) +
     geom_line(color='grey') + 
-    geom_errorbar(aes(ymin=CI.low,ymax=CI.high, color=Risk), width=5, show.legend=F) + 
+    geom_errorbar(aes(ymin=CI.low,ymax=CI.high), color='grey39', width=3, show.legend=F, size=1) + 
     geom_point(aes(color=Risk), size=5) + 
-    scale_color_manual(values=riskColors(), limits=names(riskColors())) + 
+    scale_color_manual(values=unlist(riskColors()), limits=names(riskColors())) + 
     scale_x_date(date_breaks = "2 month", date_labels = "%b %Y") +
     labs(y='Absolute Risk', x='Endoscopy Date',title='Absolute risks over time') + 
     theme_bw() + theme(legend.position='bottom', axis.text.x = element_text(angle=45, hjust=1))

R/process_swgs.R

@@ -104,7 +104,6 @@ runQDNAseq<-function(bam=NULL,path=NULL,outputPath=NULL, minMapQ=37, binsize=50)
 #' @export
 loadSampleInformation<-function(samples, path=c('NDBE','ID','LGD','HGD','IMC','OAC')) {
   if (is.character(samples)) {
-    #sample.info = .readFile(samples, col_types = cols('Patient ID'=col_character(), 'Pathology'=col_character(), 'Sample'=col_character(), 'P53 IHC'=col_integer()))
     sample.info = .readFile(samples)
   } else if (is.data.frame(samples)) {
     sample.info = samples
@@ -117,11 +116,14 @@ loadSampleInformation<-function(samples, path=c('NDBE','ID','LGD','HGD','IMC','O
 
   colnames(sample.info) = .titleCase(colnames(sample.info))
 
-  exp_cols = c('Pathology','GEJ.Distance', 'P53 IHC')
-  cols_found = sapply( exp_cols, function(x) x %in% colnames(sample.info) )
-
-  if (length(which(!cols_found)) > 0)
-    warning(paste0('Missing expected columns from sample information: ',paste(names(which(!cols_found)), collapse=', '), ". Recommendations will be based on predicted risks and: ",paste(names(which(cols_found)), collapse=', ')) )
+  col_regex = 'Pathology|GEJ(\\.| )Distance|P53(\\.| )IHC'
+  cols_found = grep(col_regex, colnames(sample.info), value=T, ignore.case = T)
+
+  if (length(cols_found) < 3) {
+    message = paste0("Missing expected columns from sample information. Recommendations will be based on predicted risks")
+    if (length(cols_found) > 0) message = paste0(message, ' and: ', paste(cols_found,collapse=','))
+    warning(message)
+  }
 
   endo.date<-function(endo) {
     if (is.character(endo)) {
@@ -136,18 +138,40 @@ loadSampleInformation<-function(samples, path=c('NDBE','ID','LGD','HGD','IMC','O
       as.Date(endo)
     }
   }
-
-  sample.info = sample.info %>% rowwise %>% dplyr::mutate(Endoscopy = endo.date(Endoscopy))
 
-  #sample.info$Endoscopy = factor(sample.info$Endoscopy, ordered=T)
+  # Don't change the factor if it's already done.
+  gej_col = grep('GEJ(\\.| )Distance', colnames(sample.info), value=T)
+  if (length(gej_col) > 0 && !is.factor(sample.info[[gej_col]])) {
+    gej.dist<-function(gej) {
+        funct = case_when(
+          length(which(is.na(as.numeric(gej)))) <= 0 ~ 'as.numeric',
+          length(which(is.na(as.character(gej)))) <= 0 ~ 'as.character'
+        )
+        sapply(gej, eval(funct))
+      }
 
-  pathCol = grep('Pathology',colnames(sample.info))
-  if (length(pathCol) > 0) 
-    sample.info[[pathCol]] = factor(sample.info[[pathCol]], levels=path, labels=path, ordered=T)
+    sample.info = sample.info %>% 
+      dplyr::mutate_at(dplyr::vars(!!gej_col), list(~gej.dist(.)))  
+
+    levels = sort(sample.info %>% dplyr::select(!!gej_col) %>% distinct %>% pull)
 
-  p53Col = grep('P53',colnames(sample.info))
-  if (length(p53Col) > 0) 
-    sample.info[[p53Col]] = factor(sample.info[[p53Col]], levels=c(0,1), labels=c('Normal','Aberrant'), ordered=T)
+    sample.info = sample.info %>% 
+      dplyr::mutate_at(dplyr::vars(!!gej_col), list(~factor(., levels=levels, ordered=T)))
+  }  
+
+  p53_col = grep('P53', colnames(sample.info), value=T)
+  if (length(p53_col) > 0 && 
+      length(grep('Normal|Aberrant', dplyr::select(sample.info, !!p53_col) %>% distinct %>% pull)) > 0) {
+    sample.info = sample.info %>% 
+      dplyr::mutate_at(dplyr::vars(matches('P53')), list(~factor(., levels=c('Normal','Aberrant'), ordered=T)))
+  } else {
+    sample.info = sample.info %>% dplyr::mutate_at(dplyr::vars(dplyr::matches('P53')), list(~factor(., levels=c(0,1), labels=c('Normal','Aberrant'), ordered=T)))
+  }
+
+  sample.info = sample.info %>% rowwise %>% dplyr::mutate(Endoscopy = endo.date(Endoscopy))
+
+  sample.info = sample.info %>% 
+    dplyr::mutate_at(dplyr::vars(dplyr::matches('Pathology')), list(~factor(., levels=path, ordered=T))) 
 
   class(sample.info) <- c('SampleInformation', class(sample.info))
   return(sample.info)

R/risk_prediction.R

@@ -109,7 +109,7 @@ per.sample.error<-function(df, seg.tile.error, arm.tile.error, be.model) {
 }
 
 
-predictRisk<-function(obj, merged.tiles, be.model = NULL, verbose=T) {
+predictRisk<-function(obj, merged.tiles, be.model=NULL, verbose=T) {
   if (is.null(be.model)) {
     be.model = BarrettsProgressionRisk:::be_model
   } 
@@ -119,7 +119,8 @@ predictRisk<-function(obj, merged.tiles, be.model = NULL, verbose=T) {
   for(i in colnames(merged.tiles$tiles)) sparsed_test_data[,i] = merged.tiles$tiles[,i]
 
   perSampleError = tibble('Sample'=names(merged.tiles$per.sample.error),'Error'=merged.tiles$per.sample.error)
-  perSampleError = left_join(perSampleError, obj$sample.info, by='Sample') %>% dplyr::select('Sample','Error','Endoscopy')
+  perSampleError = left_join(perSampleError, obj$sample.info, by='Sample') %>% 
+    dplyr::select('Sample','Error','Endoscopy')
 
   # Predict and generate absolute probabilities
   RR = predict(be.model$fit, newx=sparsed_test_data, s=be.model$lambda, type='link')
@@ -166,7 +167,7 @@ predictRiskFromSegments<-function(obj, be.model = NULL, verbose=T) {
 
   # Tile, scale, then merge segmented values into 5Mb and arm-length windows across the genome.
   binnedSamples = tryCatch({
-    tileSamples(obj, be.model, verbose)
+    tileSamples(obj, be.model, scale=T, MARGIN=2, verbose=verbose)
   }, error = function(e) {
     msg = paste("ERROR tiling segmented data:", e)
     stop(msg)
@@ -375,14 +376,16 @@ rx<-function(brr, by=c('endoscopy','sample')) {
   time = match.arg(time)
 
   if (riskBy == 'Sample') warning('Rx rules are intended to be applied on a per-endoscopy basis, not per-sample.')
-
-  if (!is.numeric.Date(preds[[riskBy]]) | !is.numeric(preds[[riskBy]])) {
+
+  time = 'date'  
+  if (!class(preds[[riskBy]]) %in% 'Date' || is.numeric(preds[[riskBy]])) {
     time = 'numeric'
     preds[[riskBy]] = factor(preds[[riskBy]])
   }
 
-  preds = preds %>% rowwise() %>% dplyr::mutate(Risk = .risk(Probability, pred.confidence)) %>%
-    mutate(Risk = factor(Risk, levels=c('Low','Moderate','High'), ordered=T))
+  preds = preds %>% rowwise() %>%
+    dplyr::mutate(Risk = .risk(Probability, pred.confidence)) %>%
+    mutate(Risk = factor(Risk, levels=names(riskColors()), ordered=T))
 
   p53Col = grep('p53', colnames(preds), value=T, ignore.case=T)
   pathCol = grep('pathology', colnames(preds), value=T, ignore.case=T)
@@ -432,10 +435,14 @@ rx<-function(brr, by=c('endoscopy','sample')) {
       rules = add_row(rules, 'Time 1' = t1, 'Time 2' = t2, 'Rule' = rule) 
     }
 
+    # If the last pair resulted in an increased surveillance rx than this one should do as well?
+    if ( i == nrow(preds) && rules[i,'Rule'] > rules[(i-1), 'Rule'] ) rules[i,'Rule'] = rules[(i-1), 'Rule'] 
+
     if (i == nrow(preds)) break;
   }
   rules = rules %>% mutate(Rx = map_chr(Rule, .rule.rx))
 
+
   return(rules)
 }
 

R/utility_functions.R

@@ -48,7 +48,7 @@ getcachedir<-function() {
 }
 
 # Renames 'get.chr.lengths'
-chrInfo<-function(chrs=c(1:22, 'X','Y'), prefix='chr', build='hg19', file=NULL) {
+chrInfo<-function(chrs=c(1:22, 'X','Y'), prefix='chr', build='hg19', file=NULL, verbose=F) {
   local_file =  paste(build,'_info.txt',sep='')
 
   local_file = system.file("extdata", local_file, package="BarrettsProgressionRisk")
@@ -60,7 +60,7 @@ chrInfo<-function(chrs=c(1:22, 'X','Y'), prefix='chr', build='hg19', file=NULL)
     file = tmp_file
   }
 
-  if (!is.null(file) && file.exists(file)) {
+  if (!is.null(file) && file.exists(file) && verbose) {
     message(paste0("Reading chromosome information for build ",build," from ",file))
 
     chr.lengths = read.table(file, header = T, sep='\t', colClasses = c(character(), numeric(), numeric(), numeric(), numeric()), stringsAsFactors = F) %>% as_tibble()

README.md

@@ -15,50 +15,57 @@ library(BarrettsProgressionRisk)
 # Load example data set
 data(package='BarrettsProgressionRisk',ExampleQDNAseqData)
 
+# bamPath needs to contain one or more bam files for a patient, the default binsize is 50. 
+# This should not be changed without extensive testing of the data unless you are retraining the underlying model!  
+
+# example of fitted data output from QDNAseq function call: runQDNAseq(bamPath='.', outputPath='<my path>/', binsize=50)
 print(fit.data)
+# example of raw data output from QDNAseq function call
 print(raw.data)
+
+# example of sample information table
 print(info)
 
+# Segment the fitted and raw data
 segObj = segmentRawData(loadSampleInformation(info), raw.data, fit.data, verbose=F) 
 
+# Predict risks from segmented data
 pr = predictRiskFromSegments(segObj, verbose=F)
 
-## Results
+## -- Results  -- ##
 
 # get QC information
 sampleResiduals(pr)
 
 # plot raw data
 plotSegmentData(pr)
 
-# risk per sample
-predictions(pr,'sample')
-
 # Per sample mountain plots with annotations for coefficients
 copyNumberMountainPlot(pr, annotate=T)
 
+### Per sample classifications/probabilities
 
-# TODO error
-patientRiskTilesPlot(pr)
+# risk per sample for samples that pass QC (see sampleResiduals(...))
+predictions(pr,'sample')
 
-# TODO error
-patientEndoscopyPlot(pr)
+# output the absolute risk CI per sample
+absoluteRiskCI(pr, 'sample')
 
+# Plot the sample classifications by Endoscopy and location 
+patientRiskTilesPlot(pr)
+
+### Per endoscopy classifications/probabilities
 
 # risk per endoscopy
 predictions(pr,'endoscopy')
 
-# output the absolute risk CI per sample
-absoluteRiskCI(pr, 'sample')
-
 # output the absolute risk CI per endoscopy
 absoluteRiskCI(pr, 'endoscopy')
 
-
-
+# Plot the endoscopy predictions over time, with risk classifications and confidence intervals based on the table output by absoluteRiskCI(...)
+patientEndoscopyPlot(pr)
 
 # Get recommendations per endoscopy, given as either sequential integers or dates in the sample information loaded initially.
-
 rx(pr)
 
 ```
@@ -77,14 +84,11 @@ library(knitr)
 
 # bamPath needs to contain one or more bam files for a patient, the default binsize is 50. 
 # This should not be changed without extensive testing of the data unless you are retraining the underlying model!  
-BarrettsProgressionRisk::runQDNAseq(bamPath='.', outputPath=<path to qdnaseq output>,  binsize=50)
-
+# BarrettsProgressionRisk::runQDNAseq(bamPath='.', outputPath=<path to qdnaseq output>,  binsize=50)
 
-# qdnaseq.path=<path to qdnaseq output>
-qdnaseq.path='examples/'
-# info.file=<path to per sample p53 IHC/pathology file>
-info.file = 'example/endoscopy.xlsx'
 
+qdnaseq.path=system.file('extdata/example',package="BarrettsProgressionRisk")
+info.file = system.file('extdata/example','endoscopy.xlsx',package="BarrettsProgressionRisk")
 output.dir='~/tmp'
 
 options(warn = -1)