search

isaacovercast / fsc2_scripts

Shell scripts for automating replicate runs of fastsimcoal2
4 stars 0 forks source link

bootstrapping in fsc27 #1

Open grcvhon opened 1 year ago

grcvhon commented 1 year ago

Hi Isaac,

I am having an issue about bootstrapping in fastsimcoal2 and appears that my point estimates fall outside 95% confidence intervals.

I have tried non-parametric bootstrapping by sampling with replacement my VCF (thanks for the modifications in the easySFS.py script specifically --resample-vcf) but after many attempts, my point estimates from observed data still fall outside the 95%CIs.

I may be doing something that's totally wrong and would like to use your fsc2_scripts for bootstrapping however I was having difficulty interpreting how to use them. I was hoping if you could give some direction.

Thanks so much!

-Vhon

isaacovercast commented 1 year ago

Hello Vhon, Can you attach your par, est, tpl and one or more SFS files? I will be very busy next week, so will try to look at this then.

grcvhon commented 1 year ago

Thanks Isaac!

I have attached the files and also provide description for each in README.txt (included in the zip file). git-bootstrap_fsc.zip

The ways I've tried non-parametric bootstrapping to obtain 95%CI included: 1) simulating 100 SFSs then performing 20 independent runs (using -n coalescent simulations and -L optimisation cycles: -n1000000 -L100 and -n100000 -L30 [this one is mainly to speed up and from fsc27 manual too]) 2) simulating 100 SFSs, supplying --initValues file.pv (as in fsc27 manual), then performing 20 independent runs 3) simulating 1000 SFSs and running fsc2 with --initValues file.pv

But for some reason, my point estimates are always outside of the 95%CI. For example: Point estimate for effective population size parameter SEMP = 644,356 95%CI obtained using bootstrapping number (2) above: 659,668.4 - 681,601.9

Many thanks again!