Identify and create a new and public (large-ish) multiassay dataset.

[lianos]

We'll need to have this in place to more easily showcase the FacileData API.

Some choices could include:

• The ROSMAP project. This may be ideal since it has multiple assays. The caveat is that I'm not sure if the data processed (quantitated) data can be made. This has (at least) these following assays of interest:
  • RNA-seq (including microglia specific RNA-seq)
  • miRNA-array
  • metabolomics
  • Sooo awesome!
• ARCHS4. It's only rnaseq, unfortunately, but can use gene-level and transcript-level quantitation as different assays.
• A FacileCCLEDataSet, which includes:
  • Gene-level RNA-seq from the cell lines
  • isoform-level RNA-seq?
  • achiles 2.0/ceres scores as another assay
  • "hotspot" mutations for oncogenes (and gross-level tumor suppressor hits) as sample covariets
  • We can also explore how one might define genesets in here based on the recent L1000 uber cmap data ... borrowing ideas from loom's "feature edges" might be worth thinking about (for the future)
• We can always fall back to doing the TCGA, but am slightly biased against this since the FacileTCGADataSet is floating around ...
• @phaverty and @VRouilly, any other ideas?

[lianos]

I guess CCLE is the only option, we can get:

1. Gene-level expression
   • From MSSM/Harmonize
   • I know I have another copy of this somewhere
2. Transcript-level expression (where?)
   • I know I've seen this already processed from somewhere, as well.
3. Gene-level CNV values from MSSM/Harmonizome
4. Gene-level mutation calls MSSM/Harmonizome
5. Gene essentiality / achilles

[phaverty]

I’ve got a script to organize all the CCLE stuff. Will share.

On Wed, Sep 26, 2018 at 3:17 PM Steve Lianoglou notifications@github.com wrote:

I guess CCLE is the only option, we can get:

1. Gene-level expression (where?)
2. Transcript-level expression (where?)
3. Gene-level CNV values from MSSM/Harmonizome http://amp.pharm.mssm.edu/Harmonizome/dataset/CCLE+Cell+Line+Gene+CNV+Profiles

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/Genentech/FacileDataSet/issues/12#issuecomment-424887850, or mute the thread https://github.com/notifications/unsubscribe-auth/AH02KyUazseqnyP_iL5O1BnGAazyytNhks5ue_z3gaJpZM4WIMUL .

-- Pete

---

Peter M. Haverty, Ph.D. Genentech, Inc. phaverty@gene.com

[phaverty]

I use these files CCLE_copynumber_byGene_2013-12-03.txt CCLE_RNAseq_081117.rpkm.gct ccle2maf_081117.txt

from here: https://portals.broadinstitute.org/ccle/data

' Build long-form table of CCLE RPKM values

'

' Does conversion from CCLE Ensembl IDs to Entrez

' @param ccle_expr_file single character, *rpkm.gct file from CCLE FTP

' @return data.frame with ccle_name, gene_id and rpkm columns

' @export

build_ccle_expr <- function(ccle_expr_file) { expr = read_tsv(ccle_expr_file, skip = 2) colnames(expr)[1] = "Xref" expr$Xref = gsub("\.\d+$", "", expr$Xref) expr[[2]] = NULL expr_ginfo = translate(expr$Xref, org.Hs.egENSEMBL2EG, return.list = FALSE) expr_ginfo$to = paste0("GeneID:", expr_ginfo$to) colnames(expr_ginfo) = c("gene_id", "Xref") expr_ginfo = expr_ginfo[order(as.integer(gsub("ENSG", "", expr_ginfo$Xref))), ] expr_ginfo = expr_ginfo[!duplicated(expr_ginfo$gene_id), ] expr_ginfo = expr_ginfo[!duplicated(expr_ginfo$Xref), ] stopifnot(length(setdiff(expr_ginfo$Xref, expr$Xref)) == 0) stopifnot(!anyDuplicated(expr$Xref)) expr = inner_join(expr_ginfo, expr, by = "Xref") expr$Xref = NULL expr_long = melt(expr, id.vars = c("gene_id"), variable.name = "ccle_name", value.name = "rpkm") expr_long }

' Build long-form table of CCLE relative copy number values

'

' Build long-form table of CCLE relative copy number values

' @param ccle_cn_file single character, e.g.

CCLE_copynumber_byGene_2013-12-03.txt from CCLE FTP

' @return data.frame with ccle_name, gene_id and relative_cn (log2ratio)

columns

' @export

build_ccle_cn <- function(ccle_cn_file) { cn = read_tsv(ccle_cn_file) cn = dplyr::select(cn, -c(SYMBOL, CHR, CHRLOC, CHRLOCEND)) cn = dplyr::rename(cn, gene_id = EGID) cn$gene_id = paste0("GeneID:", cn$gene_id) cn_long = melt(cn, id.vars = "gene_id", variable.name = "ccle_name", value.name = "relative_cn") cn_long }

' Build table of mutation details for CCLE lines

'

' Build table of mutation details for CCLE lines

' @param ccle_mut_file single character, e.g. ccle2maf_081117.txt from

CCLE FTP

' @param genes data.frame, from build_genes_table

' @param ccle_sinfo data.frame, from sinfo_from_ceres

' @return data.frame

' @export

build_ccle_fullgene <- function(ccle_mut_file, genes, ccle_sinfo) { fullgene = read_tsv(ccle_mut_file, col_types = "cccciicccccccccccccllilidccccccc") fullgene = dplyr::select(fullgene, -Hugo_Symbol) fullgene = dplyr::rename(fullgene, gene_id = Entrez_Gene_Id, ccle_name = Tumor_Sample_Barcode) fullgene = filter(fullgene, gene_id != 0) # WTF, Broad? Other id/symbol pairs seem ok

fullgene$gene_id = paste0("GeneID:", fullgene$gene_id)

## Add allele frequency, allele counts seem to be ALT:REF as there are

lots of X:0

## and no 0:X x = group_by(fullgene, ccle_name) foo = summarize(x, WES

=

## all(is.na(WES_AC)), WGS = all(is.na(WGS_AC)), RNA =

all(is.na(RNAseq_AC)))

## variants appear to mostly be from RNASeq

fullgene$joint_AC = ifelse(!is.na(fullgene$WGS_AC), fullgene$WGS_AC,

ifelse(!is.na(fullgene$WES_AC),

fullgene$WES_AC, ifelse(!is.na(fullgene$RNAseq_AC), fullgene$RNAseq_AC,

fullgene$SangerRecalibWES_AC))) fullgene$altcount = as.integer(gsub("^(\d+):.$", "\1", fullgene$joint_AC)) fullgene$refcount = as.integer(gsub("^.:(\d+)$", "\1", fullgene$joint_AC)) fullgene$variant_frequency = fullgene$altcount/(fullgene$altcount + fullgene$refcount) fullgene$altcount = fullgene$refcount = fullgene$joint_AC = NULL

## Assign hotspot and lof status

fullgene = mutate(fullgene, is_hot = isTCGAhotspot | isCOSMIChotspot)
fullgene = mutate(fullgene, is_lof = is_hot | (isDeleterious & (ExAC_AF

< 0.01 | is.na(ExAC_AF))))

## Join in gene_symbol and tissue

fullgene = left_join(dplyr::select(genes, gene_id, gene_symbol),

fullgene, by = "gene_id") fullgene = left_join(dplyr::select(ccle_sinfo, ccle_name, cell_line, tissue), fullgene, by = "ccle_name") fullgene = dplyr::select(fullgene, gene_id, gene_symbol, ccle_name, cell_line, tissue, everything())

fullgene

}

' Build long-form summary table of gene mutation status

'

' Contains loss-of-function and hotspot annotations per sample and gene

' @param fullgene data.frame from build_ccle_fullgene()

' @return data.frame

' @export

build_ccle_mut <- function(fullgene) {

Make lof and hotspot tables

## -Inf varfreq becomes 0

lof_long = dplyr::summarize(group_by(filter(fullgene, is_lof), gene_id,

ccle_name), max_lof_varfreq = max(variant_frequency, na.rm = TRUE))

hot_long = dplyr::summarize(group_by(filter(fullgene, is_hot), gene_id,

ccle_name), max_hot_varfreq = max(variant_frequency, na.rm = TRUE))

mut_long = full_join(lof_long, hot_long, by = c("gene_id", "ccle_name"))
mut_long = mutate(
    mut_long,
    max_lof_varfreq = ifelse(is.infinite(max_lof_varfreq), 0,

max_lof_varfreq), max_hot_varfreq= ifelse(is.infinite(max_hot_varfreq), 0, max_hot_varfreq) ) mut_long }

' Join ceres, expression, copy number and mutation status tables

'

' Join ceres, expression, copy number and mutation status tables

' @param ceres_long data.frame, from build_ceres_table

' @param expr_long data.frame, from build_ccle_expr

' @param cn_long data.frame, from build_ccle_cn

' @param mut_long data.frame, from build_ccle_mut

' @param genes data.frame, from build_genes_table or

add_gene_drugability

' @param ccle_sinfo data.frame, from sinfo_from_ceres

' @return data.frame

' @export

build_ccle_genomics <- function(ceres_long, expr_long, cn_long, mut_long, genes, ccle_sinfo) {

expr_long = filter(expr_long, ccle_name %in% ccle_sinfo$ccle_name)
cn_long = filter(cn_long, ccle_name %in% ccle_sinfo$ccle_name)
mut_long = filter(mut_long, ccle_name %in% ccle_sinfo$ccle_name)

stopifnot(!anyDuplicated(ceres_long[, c("gene_id", "ccle_name")]))
stopifnot(!anyDuplicated(expr_long[, c("gene_id", "ccle_name")]))
stopifnot(!anyDuplicated(cn_long[, c("gene_id", "ccle_name")]))
stopifnot(!anyDuplicated(mut_long[, c("gene_id", "ccle_name")]))

genomics_long = full_join(expr_long, cn_long, by = c("gene_id",

"ccle_name")) genomics_long = full_join(genomics_long, mut_long, by = c("gene_id", "ccle_name"))

lines_missing_genomics = setdiff(ceres_long$ccle_name,

genomics_long$ccle_name)

write.csv(lines_missing_genomics, file =

"broad_lines_with_ceres_not_ccle.csv")

genomics_long = genomics_long[genomics_long$gene_id %in%

ceres_long$gene_id, ] genomics_long = full_join(genomics_long, ceres_long, by = c("gene_id", "ccle_name")) genomics_long = left_join(dplyr::select(genes, gene_symbol, gene_id), genomics_long, by = "gene_id") genomics_long = left_join(dplyr::select(ccle_sinfo, ccle_name, cell_line, tissue), genomics_long, by = "ccle_name") genomics_long = dplyr::select(genomics_long, gene_id, gene_symbol, ccle_name, cell_line, tissue, everything())

genomics_long

}

Pete

---

Peter M. Haverty, Ph.D. Genentech, Inc. phaverty@gene.com

On Wed, Sep 26, 2018 at 5:31 PM, Pete Haverty phaverty@gene.com wrote:

I’ve got a script to organize all the CCLE stuff. Will share.

On Wed, Sep 26, 2018 at 3:17 PM Steve Lianoglou notifications@github.com wrote:

I guess CCLE is the only option, we can get:

1. Gene-level expression (where?)
2. Transcript-level expression (where?)
3. Gene-level CNV values from MSSM/Harmonizome http://amp.pharm.mssm.edu/Harmonizome/dataset/CCLE+Cell+Line+Gene+CNV+Profiles

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/Genentech/FacileDataSet/issues/12#issuecomment-424887850, or mute the thread https://github.com/notifications/unsubscribe-auth/AH02KyUazseqnyP_iL5O1BnGAazyytNhks5ue_z3gaJpZM4WIMUL .

-- Pete

---

Peter M. Haverty, Ph.D. Genentech, Inc. phaverty@gene.com

[lianos]

Thanks, Pete: this should prove very helpful!