Copy Number Variation

This is a repo to compile and harmonizr various versions of CNV code that was originally used for mad4hatter. Should be made more flexible to allow for panel/ target alterations and incorporate various updates to method people have made.

Notes from meeting with Andres in Barcelona

Currently

• Current repo https://github.com/andres-ad/madh_utilities
  • Now forked and code is under madh_utilities_estCNV/
• Rmd needs to be updated
• Loci info is for MH panel, theres a version for PfPHAST andres needs to push
• Includes manifest example
• PCA no one uses - was intended to be a clustering of runs into batches
  • Not been properly benchmarked etc.

In andres pfphast rmd

• Need to set paths manually
  • allele data
  • manifest
  • locus info
  • thresholds
    • new ones for PfPHAST compared to MH
• Locus info file (RDS) includes
  • Amplicon info
  • Locus filter - good for CNV. ones that amplify well
  • Loci of interest
    • Target (e.g. mdr1) and loci to use for it
• Manifest
  • required : SampleID, Batch, SuperBatch, CNVControl (non duplicates)
• Code
  • Loads everything
  • Remove anything that’s not a Pf target
  • Fit a Gamm, use residuals to normalise everything and find the DCR
• Outputs
  • Html with all the code
  • Some QC
    • coverage
    • From the manifest - how many samples and controls are in the superbatch
    • Plots Gamm for controls so you can see what it looks like
    • For controls what is observed and black x is the median to normalise to
      • Sanity check here that there isn’t something different happening between controls within a locus and there aren’t different profiles happening for batches across loci
    • Use this to normalise and group loci (fitting GAMM) to estimate fold change of samples
    • HRP2 is fitted separately because of different breakpoints
      • 2 targets land on HRP, 2 targets are downstream of it
      • Ones on hrp2 underperform - one is excluded from analysis
      • Both targets sitting on HRP2 and ones downstream need to be deleted to be classified as a deletion.
        • Fit separately and take a maximum to decide
    • Then final results are show with HRP2 collapsed as one target
  • Fold change outputs for each gene

Yos

• Took the one on github that Andres had adapted to take v0.1.8 outputs
• Made R script callable on command line
• Unsure what parasitemia is being used for
• Can change thresholds and set outputs on command line

If people don’t have controls

• Assume majority of samples have no CNV and use the median of the data for normalisation
• This code is not in repo - Jessica and Andres have that
• This should be added as a feature
• This was found to work better (comments on slack from Bryan though in immrse channel)