Introduction

• Expand on bread wheat genome challenges
  • polyploid
  • repeats

Existing aligners

how detailed do we need to be? different alignment approaches. cover pseudo alignment? ie kallisto, salmon?

Methods

• Reference genome/s (probably one is enough?)
  • Zimin/Salzberg
  • Clavijo
  • one would be enough, but in a pipeline there should be none or very little overhead of having 2
  • can we quantify 'complexity'? ie repeats/duplications/syntenic regions with limited polymorphisms... ideally from existing work)
    • Worth doing even though this can be only indicative - really long (almost identical) repeats will be missing from the assemblies!
    • would any of Stuart's hamming stats be of use?
    • k-mer frequency analysis either on the assembly or some raw data (either CS or RAC8754 PE+MP) - should be quick, with KMC can go up to read length (max k=255). When a plot illustrating that includes various species, it can be very informative
    • genome mappability score or similar
• Align: (what parameter space? can't be too complete)
  • DNA reads (bioplatforms? - selected highest coverage cultivars?)
  • simulated reads
    • can we use biokanga simreads?
    • and/or some third-party?
    • Parameter space to be explored: error rates should be close to reported empirical
  • alignment parameters e.g. max substitutions
    • could go slightly outside the range used when aligning reads from/to wheat
    • soft clipping / chimeric trimming and their influence on speed, align rate, false positive rate...
    • PE or SE? If PE do we need to explore the various PE modes, how much can they contribute?
• Measure:
  • Standard parameters (time/cores blah blah)
  • %aligned in different classes (unaligned, unique etc)
  • Accuracy for simulated reads
  • SNP calling? Typically this is done with a different tool to the aligner - are we introducing SNP calling as well? unsure. could work well with bioplatforms data.

Simulated reads procedure

Much of this can be taken care of by using https://github.com/karel-brinda/rnftools

for each assembly #may be informative to show arabidopsis, rice, barley, wild emmer wheat, bread wheat 
  for each read_length #fix?
    for each err rate 
      simulate n-fold coverage reads (proportional to assembly/genome size) 
      for each param settings in {default, ..., ...} 
        align reads to ref using each tool
        quantify speed, accuracy

Results

Simulated reads

I can imagine a figure showing the fate of reads under different aligners. A sort of snakey diagram showing where they end up, and whether it's the best place or not.

bioplatforms reads

Summary of different SNPs reported (ie what is the overlap in the SNPs identified using the different aligners - would need to use the same SNPs caller on the BAM output I guess)? I don't actually know what the differences would look like. Would be nice to have an example or two of where a difference is. Maybe this is really down to parameters (ie how many reads you need to trust the call?)

Discussion