1 - Infer ploidy of X and Y - model typical situations of read balance for reference comparison

[mathbionerd]

Simulate, or generate from real data, a reference distribution of (a) depth and (b) read balance on X and Y vs autosomes for (where appropriate):

• XX

• XY

• X0

• XXY

• XYY

• XXX

• XXXY

• XXXX

• XXXXY

• XXYY

  Typical expectations for heterozygous calls under different sex chromosome complements:

  Genotype	X_call	Y_call
  XX	het	none
  XY	hap	hap
  X0	hap	none or partial_hap
  XXY	het or hap	hap
  XYY	hap	hap
  XXX	het	none
  XXXY	het/hap	hap
  XXXX	het/hap	none
  XXXXY	het/hap	hap
  XXYY	het/hap	hap

  Expectations for depth under different sex chromosome complements relative to diploid autosomes:

  Genotype	X_depth/A_depth	Y_depth/A_depth
  XX	1x	0x
  XY	0.5x	1x
  X0	0.5x	0x (or partial)
  XXY	1x	0.5x
  XYY	0.5x	1x
  XXX	1.5x	0x
  XXXY	1.5x	0.5x
  XXXX	2x	0x
  XXXXY	2x	0.5x
  XXYY	1x	1x

[mathbionerd]

Considering simulating a reference dataset, to understand coverage across the genome, ideally it will be fast enough to simulate the whole genome, which would be:

46,XX:
For each autosome, simulate 10X coverage
For X, simulate 10X coverage

46,XY:
For each autosome simulate 10X coverage
For X, simulate 5X coverage
For Y, simulate 5X coverage

Alternatively, if it is really slow, we could try just four chromosomes:

46,XX:
chr 1: 10X
chr 22: 10X
chr X: 10X

46,XY:
chr 1: 10X
chr 22: 10X
chr X: 5X
chr Y: 5X

I think the approach of varying coverage, and merging should be okay, for inferring both read balance and depth:

45, X - version 1:
For each autosome, simulate 10X coverage
For X, simulate 5X coverage

45,X - version 2 (if possible):
For each autosome, simulate 10X coverage
For X, simulate 5X coverage
For Y, simulate 5X coverage of only the first 10Mb.

47,XXY:
For each autosome simulate 10X coverage
For X, simulate 10X coverage
For Y, simulate 5X coverage

Note: for about half of these cases, we expect the X to be identical, and of paternal origin, so I'm not sure if we can do one version where we simulate 10X from a diploid pair of X chromosomes, and a second where we simulation 10X from a haploid X chromosome.

[thw17]

This strategy kind will bias our data in a way that we won't be able to understand how read depth is distributed (I.e. If we say X1 has 10x and X2 has 5X - to simulate two X from one parent and one from the other - we will know exactly how much depth the X chromosome has). We need to carefully simulate an entire diploid genome with the desired numbers of sex chromosomes to really understand how reads are distributed in this setting.

[tanyaphung]

Hi Tim,

So if I simulate a diploid genome, and specifying that the X chromosome is triploid (for the 47, XXX case), do you think this will avoid the bias issue that you mention?

From page 2 of this recent review http://www.nature.com/nrg/journal/v17/n8/pdf/nrg.2016.57.pdf it looks like the simulator EAGLE allows me to do that.

Thanks,
Tanya

[mathbionerd]

I thought one important point was to say, if we know how much the depth is, how bad do we do at aligning these regions when we merge all the reads and try to align? I guess that is a different question that you were getting at?

[Phillip-a-richmond]

@tnphung does EAGLE allow for variant incorporation? We could simultaneously test var-calling abilities if it does.

[mathbionerd]

@Phillip-a-richmond - can you look into EAGLE?

[tanyaphung]

@Phillip-a-richmond I think so. it has a --variant flag to apply variants to a reference genome. I will look into how to run this too.

[thw17]

@mwilsonsayres The simulations Tanya and I are working on are to support our method to infer ploidy. The goal of the simulations is - what does depth on the X and or Y look like in these various aneuploidies? This depends not simply on the numbers of X or Y, but the fraction of the total diploid sequence that is X or Y (that's my working hypothesis at least). That is, we need to understand this in the context of the whole genome.

[ekarlins]

Do we do any sort of normalization for sequencing depth before attempting to infer ploidy? Quantile normalization or the like?

This could help adjust for GC content or other biology that may effect sequencing depth.

The flip side is that if the ploidy estimate is working well without normalization it might not be worth the extra computation.

Eric

On Oct 28, 2016, at 3:13 AM, Tim Webster notifications@github.com wrote:

@mwilsonsayres The simulations Tanya and I are working on are to support our method to infer ploidy. The goal of the simulations is - what does depth on the X and or Y look like in these various aneuploidies? This depends not simply on the numbers of X or Y, but the fraction of the total diploid sequence that is X or Y (that's my working hypothesis at least). That is, we need to understand this in the context of the whole genome.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.