gregorgorjanc
commented
3 years ago Here is a comparison of MaCS (as called from AlphaSimR in R) vs msprime (as called from stdpopsim in python). Both cases implement the Macleod et al (2013) Bos Taurus (cattle) model - see https://github.com/popsim-consortium/stdpopsim/tree/main/stdpopsim/catalog/BosTau. AlphaSimR is simulating 10^8 bp chromosomes with 1 Morgan length. So this is a reasonable chromosome. AlphaSimR call to MaCS can be seen at https://github.com/gaynorr/AlphaSimR/blob/master/R/founderPop.R (150-154) - there is 10% difference in the most recent Ne between AlphaSimR and stdpopsim, but that should not cause so different timings.
# install.packages("AlphaSimR")
library(AlphaSimR)
nRep = 10
timeDiff = rep(0, times = nRep)
for (Rep in 1:nRep) {
start = Sys.time()
out = runMacs(nInd = 5, nChr = 1, species = "CATTLE") # diploid by default
end = Sys.time()
timeDiff[Rep] = end - start
print(timeDiff[Rep])
}
mean(timeDiff)
# 1.970797and
import stdpopsim
import time
import numpy as np
species = stdpopsim.get_species("BosTau")
model = species.get_demographic_model("HolsteinFriesian_1M13")
contig = species.get_contig("11")
samples = model.get_samples(10)
engine = stdpopsim.get_engine("msprime")
nRep = 1
diffTime = np.zeros(nRep)
for Rep in range(nRep):
start = time.time()
ts = engine.simulate(model, contig, samples)
end = time.time()
diffTime[Rep] = end - start
print(diffTime[Rep])
diffTime.mean()
# 503.74193811416626I am seeing even slower times with dtwf (~3900. smc and smc_prime ran for even longer.
jeromekelleher
petrelharp
Msprime is much slower than MaCS when simulating long genomes/lots of recombination, as we've seen with the BosTau model in stdpopsim.
There may be some low-hanging fruit in making this faster as this regime hasn't really been optimised for.