Parameters to vary for simulated data sets

stephenslab / dsc-log-fold-change

dsc to compare approaches to estimating/testing log-fold-change from counts

https://stephenslab.github.io/dsc-log-fold-change/

2 stars 3 forks source link

Parameters to vary for simulated data sets #3

Open jdblischak opened 5 years ago

jdblischak commented 5 years ago

We would like the simulation function vary the following parameters. Next to each parameter is a link to relevant existing code:

[x] Number of samples in each group
[x] Number of genes
[x] Independent vs. dependent inter-gene correlations (source)
[ ] Distribution of effect sizes (source)
[ ] Number of control genes (source)
[ ] Latent confounding (source)

cc @jhsiao999 @stephens999

jdblischak commented 5 years ago

Another resource is David's seqgendiff package:

This package is designed to take real RNA-seq data and alter it by adding signal to a known proportion of the genes. The advantage of this way of simulating data is that you can see how your method behaves when the simulated data exhibit common (and annoying) features of real data.

jhsiao999 commented 5 years ago

R function poisthin is currently used to generate simulated data.

Location: dsc/modules/dataSimulate.R Source: Taken from Davis' seqgendiff package Modifications:

I added signal_dist argument to specify the desired distribution for the signals, such as "big_normal" and "near_normal" (more can be added later)
I specified the distribution parameters inside the pois_thin function

Comments:

We can include this in a separate R package as the function grows