gaow / neuro-twas

Development code (private repo) for TWAS related multiomic gene-mapping in Alzheimer's disease data
6 stars 0 forks source link

Literature and methods for multivariate TWAS #9

Open gaow opened 4 years ago

gaow commented 4 years ago

@hsun3163 please fill it up.

gaow commented 4 years ago

So one point / advantage using multivariate approach is that it has more statistical power, and will boost heritability estimates so you dont "lose" too many genes. @hsun3163 could you tell from your current ROSMAP analysis roughly what proportion of genes are not usable for TWAS because of too low a heritability? And what is the heritability cutoff in FUSION?

hsun3163 commented 4 years ago

So one point / advantage using multivariate approach is that it has more statistical power, and will boost heritability estimates so you dont "lose" too many genes. @hsun3163 could you tell from your current ROSMAP analysis roughly what proportion of genes are not usable for TWAS because of too low a heritability? And what is the heritability cutoff in FUSION?

As the gene expression file is basically randomly ordered. From the first 200 genes for the gene expression file, there are 34 genes that passed the heritability check and produce a weight file. In other word, ~83% of the gene is not usable due to too low a heritability.

Currently I am rerunning the weight computation for the gene list that extracted from the Alzheimer paper with the blsmm model, which may take some time. I feel like among those genes what are the proportion of gene lost will be a more important question.

gaow commented 4 years ago

I feel like among those genes what are the proportion of gene lost will be a more important question.

Agreed. But what hertability cutoff in FUSION did you use? For those genes you can try to lower the cutoff see what you find.

gaow commented 4 years ago

Here is a recent one https://www.nature.com/articles/s41588-020-0706-2