Calculating linkage disequilibrium (LD) from Pool-seq data
- Pool sequencing data does not contain individual genotype data and hence the linear combinations of alleles per genotype cannot be known. In other words haplotype information is lost.
- What we know are allele frequencies per pool
- We have linear allele combination information per read which is limited by the sequencing technology used, i.e. for the hugely popular and currently state-of-the-art short-read fluorescence-based sequencing technology by Illumina, this is limited to 150bp tp 300bp, and an impressive ~2Mbp using nanopore sequencing.
- Considering short-read sequencing, should we look for overlapping sequences (which can be very rare if they exist at all) to somehow piece-together a longer sequence. This approach will not work for deliberately sparse genome coverage approches such as genotyping-by-sequencing (GBS) and restriction site-associated DNA sequencing (RADseq).
- Tests using LDx which does computes LD per sequencing read and therefore severely limited by maximum read length; shows that wthin this short range of 0-300 bp, the decay of LD is not yet evident (see scatterplots below).
"FALSE POSITIVE RATE" misdefined
I've misdefined "FALSE POSITIVE RATE" in my GPAS resource allocation optimisation scripts. I'm also coining the word "misdefined" to mean incorrectly attributing a word with a concept but the concept is still kind of novel and useful*.
In statistics False Positive Rate also known as Type I error rate defined as the probability of rejecting the null hypothesis given that the null hypothesis is true. However, in my GPAS resource allocation optimisation I used "False Positive Rate" to mean the probability of the null hypothesis given that the null hypothesis is rejected. In GWAS terms, I calculated it as the number of non-causal loci above the significance threshold divided by the total number of loci above the significance threshold. (Refer to the table on the left)
Whether this concept of:
P(null | reject null) = non-causal loci above threshold / total loci above threshold
is useful remain to be seen.
Looking through a Bayesian lens:
P(null | reject null)
= P(reject null | null) * P(null) / P(reject null)
= (false positive rate) * ((nLoci-nCausal)/nLoci) / (nLoci above significance threshold)
We can see that our weird GWAS metric
P(null | reject null) is positively proportional to the real false positive rate, but will increase with decreasing number of causal loci.
*Note: This sentence is a joke. I am not really coining this non-existent word.
Genome-wide association is a recent development in our search for solutions to the central problem in genetics: elucidation of the genetic underpinnings of phenotypes. And interest in it is very stable at the time of this writing.
Modelling resistance traits across a landscape using different interpolation methods
Simple linear interpolation via
interp (akima package) was performed and produced the plot on the left with yellow (low resistance) to red (high resistance) heatmap.
Smooth interpolation via generalized additive modelling:
gam (gmcv package) was also performed and produced the graph on the bottom right. However do the spline- and tensor-based GAM interpolation need to account for physical boundaries to get more sensible resistance gradient plots? And can we acquire more data points to improve the gradient map?