hello, I have some question about Struct LMM

[dkssud24]

alldata.xlsx
bedbimfam.zip
cov_agesex.txt
env_bmigroup.txt
pheno_glucose.txt

hi @horta, I am a Master degree in Kyung Hee University in Korea. I am majoring in bio informatics. I was impressed with your paper and struct LMM program. I think the idea about Struct LMM is the best.
So I tried to assign it to part of our data to your program, and then I want to make sure it worked properly.

1. Why do I get an error if I do not norm or gausinaize Pheno data?
2. Is the outcome of the assignment of Pheno, cov, and env equal to your intentions?

outcome :
chrom snp cm pos a0 a1 i pv_int pv
1 rs17106184 0.0 50909985 A G 0 0.116081 0.039977

`import os
import pandas as pd
import scipy as sp
from limix_core.util.preprocess import gaussianize
from struct_lmm import run_structlmm
from struct_lmm.utils.sugar_utils import norm_env_matrix
from pandas_plink import read_plink
import geno_sugar as gs

if name == "main":

# import genotype file
#bedfile = "data_structlmm/chrom22_subsample20_maf0.10"
bedfile = "rs17106184_N50"
(bim, fam, G) = read_plink(bedfile)

# subsample snps
#Isnp = gs.is_in(bim, ("22", 17500000, 18000000))
#G, bim = gs.snp_query(G, bim, Isnp)

# load phenotype file
phenofile = "pheno_glucose.txt"
dfp = sp.loadtxt(phenofile)
pheno = norm_env_matrix(dfp)
#dfp = pd.read_csv(phenofile, index_col=0)
#pheno = gaussianize(dfp.loc["BMI"].values[:, None])

# load environment file and normalize
envfile = "env_bmigroup.txt"
E = sp.loadtxt(envfile)
E = norm_env_matrix(E)

# mean as fixed effect
#covs = sp.ones((E.shape[0], 1))
covs = "cov_agesex.txt"
covs = sp.loadtxt(covs)

# run analysis with struct lmm
snp_preproc = {"max_miss": 0.01, "min_maf": 0.02}
res = run_structlmm(
    G, bim, pheno, E, covs=covs, batch_size=100, snp_preproc=snp_preproc
)

# export
print("Export")
print(res)
#if not os.path.exists("out"):
#    os.makedirs("out")
#res.to_csv("out/res_structlmm.csv", index=False)

`

[horta]

Hi @dkssud24 . Have you tried the new version?

Regarding question 1., struct-lmm might fail to run if your phenotype have extreme values (is that the case?). Gaussianizing the phenotype make those values become smaller.

Regarding question 2., we have updated the documentation. It is more clear now how to use the inputs. Please, have a look at https://github.com/limix/struct-lmm/blob/master/struct_lmm/_lmm.py