In brandonsie/phipcc: Generate Case-Control Reports from PhIP-seq Data

I. Introduction {.tabset}

Report Overview

This report compares the data of our case population (r phipmake::getparam(config, "proj_id"), n = r phipmake::getparam(config, "case_names") %>% strsplit(phipmake::getparam(config, "delimiter")) %>% unlist %>% length) against that of a control population (r phipmake::getparam(config, "ctrl_id"), n = r phipmake::getparam(config, "ctrl_names") %>% strsplit(phipmake::getparam(config, "delimiter")) %>% unlist %>% length) to generate a list of candidate antigens from the r phipmake::getparam(config, "library") library.

For each of these samples, we performed PhIP-Seq to measure antibody response against our libraries of phage displayed peptides.

Types of summary statistics used in this analysis

• Z-score describes normalized enrichment based on next-gen sequencing reads. Z > 10 suggests a "hit".
• RCP (Relative Control Percentile) is the fraction of controls that score lower than a given case for a given peptide/protein.

Navigating this Document

• The table of contents on the left side of the document can be used to jump between sections.
• The plots and heatmaps in this document are made interactive using plotly. This means you can mousover points for more information, drag to zoom (double click to zoom out), and select/deselect data subsets in the legend.
• Sections of this report are often subdivided with tabs. Clicking on a tab will reveal the corresponding report section.
• The code used to generate this report can be viewed by expanding the Code buttons along the right side of the document. However, this document relies on pre-computation usin the phipmake and phipcc packages from Github.

Config Documentation

These are the configuration parameters that were fed into drake::make(phipcc::define_plan_case_control).

knitr::kable(config) %>%
  kableExtra::kable_styling(bootstrap_options = c("striped", "hover", "condensed"), fixed_thead = TRUE) %>% 
  kableExtra::column_spec(2, width = "10em")

Technical Documentation

sessionInfo()

II. Candidate Antigen Tables {.tabset}

[Top]

After standard post-processing of the assay sequencing data to generate enrichment scores, we perform Fisher's exact test to compare the scores of cases vs. controls for each peptide & protein. The p values from this Fisher's test are corrected using Benjamini & Hochberg method (BH).

We now take a subset of peptides/proteins for which:

• at least r config$value[config$param == "min_hits_enrichment"] cases exhibit an enrichment score (r config$value[config$param == "data_types"]) above threshold (r config$value[config$param == "hit_thresh"]),
• the cases exhibit a hit frequency higher than that of the controls,
• The larger of (i) n = r config$value[config$param == "min_hits_rcp"] or (ii) r as.numeric(config$value[config$param == "min_freq_hits_rcp"]) * 100% rounded up of cases scored above the 95th percentile of the control population,
• Fisher's exact test gives a BH-corrected p value < r config$vlaue[config$param == "pval_thresh"] comparing case and controls against our traditional data thresholds and against an RCP threshold of 0.95.

All Hits

border <- "1px solid gray"
scroll_height <- "1000px"
use_scrollbox <- FALSE

full_table <- params$candidate_table_full
if(nrow(full_table) > 17) use_scrollbox <- TRUE

if(nrow(full_table) > 0){
  full_table <- full_table %>%
      knitr::kable(
        format = "html", escape = FALSE,
        caption = "All Candidate Antigens", row.names = FALSE) %>%
      kableExtra::kable_styling(bootstrap_options = c("striped", "hover", "condensed"), fixed_thead = TRUE) %>% 
      kableExtra::add_header_above(
        c(" " = 4, "Hit Freq" = 2,"Median Hit Score" = 2," " = 1)) %>%
       kableExtra::column_spec(1,border_left = border) %>%
       kableExtra::column_spec(ncol(full_table), border_right = border) %>%
       kableExtra::column_spec(c(1:4,6,8), border_right = border)

  if(use_scrollbox){
    full_table <- full_table %>% kableExtra::scroll_box(height = scroll_height)
  }

  full_table

}

III. Exploratory Analysis {.tabset}

[Top]

We can continue to explore this data via some graphs. Note these plots are interactive. You can click on the legend to enable/disable different traces, and you can drag a box around regions of the graph to zoom in.

Hit Frequency vs. Hit Score in Cases

We define a "hit" as scoring above the 95th percentile of the control population (RCP>0.95). Among our case samples, we plot the hit frequency among cases against the median score of these hits.

plotly::ggplotly(graphs[[1]])

Median Hit Score, Cases vs. Controls

Based on our definition of hit, there are always ~5% of controls that also score as a hit for any given peptide or protein. Here we plot the median score of hits among cases against the median score of hits among controls.

plotly::ggplotly(graphs[[2]])

P-Value vs. Median Case Hit Score

As mentioned earlier, we perform a Fisher's exact test comparing the number of hits in cases vs. controls and perform a Benjamini-Hochberg correction of the p values from this test. Here we plot the negative log base 10 of these corrected p values against the median score of hits among cases.

plotly::ggplotly(graphs[[3]])

IV. Hit Clustering {.tabset}

[Top]

Clustergram

I produce a clustergram of these hits using Ward's method for agglomeration. Each row corresponds to a peptide or protein from our filtered hit table and each column corresponds to a case sample. Squares shaded in black indicate an RCP hit, meaning that the sample scored above the 95th percentile of the control population for that peptide or protein. The dendrograms from this clustering are displayed on the following tabs.

if(!is.na(params$cluster1[[1]][1])){
  clustergram_sorted <- params$cluster1[[3]]
  htext <- params$cluster1[[2]]
  ncolors <- 2
  my_palette <- colorRampPalette(c("white","black"))(n = ncolors)


 heatmaply::heatmaply(
    clustergram_sorted, custom_hovertext = htext,
    main = paste("Candidate Antigen Clustergram"),
    colors = my_palette,
    fontsize_row = 8,
    fontsize_col = 8,
    column_text_angle = 90,
    margins = c(65, 120),
    trace = "both",
    tracecol = "grey",
    Rowv = FALSE,
    Colv = FALSE,
    grid_gap = 1,
    hide_colorbar = TRUE)
}

Row (Peptide) Dendrogram

if(!is.na(params$cluster1[[1]][1])){
  row_dendrogram <- params$cluster1[[1]]$rowDendrogram
  library(plotly)

  g <- ggdendro::ggdendrogram(row_dendrogram, rotate = TRUE) + 
    ggplot2::theme(axis.text.y = ggplot2::element_text(size = 8))
  # plotly::ggplotly(g)
  g
}

Column (Sample) Dendrogram

if(!is.na(params$cluster1[[1]][1])){
  col_dendrogram <- params$cluster1[[1]]$colDendrogram

  g <- ggdendro::ggdendrogram(col_dendrogram, rotate = TRUE) + 
    ggplot2::theme(axis.text.y = ggplot2::element_text(size = 8))
  # plotly::ggplotly(g)
  g
}

V. Epitope Clustering

[Top]

epitopefindr Motifs {.tabset}

Now we call epitopefindr, an R pipeline designed to look for BLAST alignments among these filtered hit peptides. Peptides that share sequences that might be cross-reactively recognized by a single antibody are grouped, and these groups are reclustered along with any peptides that lacked significant alignments. We find evidence for the following motifs.

#print precomputed motifs
# msa_path <- "data/epitopefindr/intermediate_files/msa/"
msa_path <- params$motifs
msa_name <- paste0(msa_path, "msa.pdf")

if(file.exists(msa_name)){
  num_page <- pdftools::pdf_info(msa_name)$pages
  num_length <- nchar(num_page)

  for(i in 1:num_page){
    num_padded <- formatC(i, width = num_length, flag = "0")
    png_name <- paste0(msa_path, "msa-", num_padded, ".png")
    pdf_name <- paste0(msa_path, "msa-", num_padded, ".pdf")

    if(!file.exists(png_name)){
          pdftools::pdf_convert(pdf_name, format = "png", filenames = png_name,
                          dpi = 300,verbose = FALSE)

    }

            cat("  \n### MSA",  i, "  \n")
        for(j in 1:length(png_name)){
            cat("![](",png_name[j],")",sep="")
          # png_name[j] %>% (EBImage::readImage) %>% (EBImage::display)
        }
        cat("  \n  \n")


  }

}

Epitope-Collapsed Hit Clustering {.tabset}

Epitope Clustergram

We re-cluster, collapsing peptides for which epitopefindr detected possible cross-reactivity into single rows.

if(!is.na(params$cluster2[[1]][1])){
  clustergram_sorted <- params$cluster2[[3]]
  htext <- params$cluster2[[2]]
  ncolors <- 5
  my_palette <- colorRampPalette(c("white","black"))(n = ncolors)


   heatmaply::heatmaply(
      clustergram_sorted, custom_hovertext = htext,
      main = paste("Candidate Antigen Clustergram (Epitope-Collapsed)"),
      colors = my_palette,
      fontsize_row = 8,
      fontsize_col = 8,
      column_text_angle = 90,
      margins = c(65, 120),
      trace = "both",
      tracecol = "grey",
      Rowv = FALSE,
      Colv = FALSE,
      grid_gap = 1,
      hide_colorbar = TRUE)

}

Row (Peptide/Epitope) Clustergram

if(!is.na(params$cluster2[[1]][1])){
  row_dendrogram <- params$cluster2[[1]]$rowDendrogram

  g <- ggdendro::ggdendrogram(row_dendrogram, rotate = TRUE) + 
    ggplot2::theme(axis.text.y = ggplot2::element_text(size = 8)) 
  # plotly::ggplotly(g)
  g
}

Column (Sample) Clustergram

if(!is.na(params$cluster2[[1]][1])){
  col_dendrogram <- params$cluster2[[1]]$colDendrogram

  g <- ggdendro::ggdendrogram(col_dendrogram, rotate = TRUE) + 
    ggplot2::theme(axis.text.y = ggplot2::element_text(size = 8)) 
  # plotly::ggplotly(g)
  g
}

brandonsie/phipcc documentation built on June 2, 2020, 6:19 a.m.