INFO - Gene ENSG... has covariance close to singular

[Maciel Rodriguez]

Hi! 

I'm using SPrediXcan to predict associations between gene expression and a disease.

I'm using Elastic Net models.

Everything seems to be going well. For each model until now, more than 70% of the model's snps were used. However, this message keeps showing up for most of the genes (not all):

INFO - Gene ENSG00000283654.1 has covariance close to singular

I'll be very gratefull if someone can tell me how should I interpret this message and what does it mean for my results.

These are the parameters I used to run SPrediXcan:

# Path to SPredixcan executable
SPrediXcan="/home/koala/biotools/MetaXcan/software/SPrediXcan.py"

python3 "$SPrediXcan" \
--model_db_path ${model} \
--covariance ${covariance_path} \
--gwas_file ${gwas} \
--gwas_file_pattern "*.txt.gz" \
--snp_column rsID \
--effect_allele_column Allele2 \
--non_effect_allele_column Allele1 \
--beta_column BETA \
--pvalue_column PVAL \
--position_column BP \
--chromosome_column CHR \
--freq_column AF \
--se_column SE \
--zscore_column SCORE \
--keep_non_rsid \
--verbosity 1 \
--additional_output \
--output_file ${output_path}

Thank you so much in advance for your kind help!

[Hae Kyung Im]

That's a red flag for numerical issues. My guess is that you don't have enough SNPs in your summary statistics. 70% is a bit lower than we would like.

Try using the gwas harmonization and imputation tools to increase your SNP coverage.

Haky

--
You received this message because you are subscribed to the Google Groups "PrediXcan/MetaXcan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to predixcanmetax...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/predixcanmetaxcan/055ea6f9-4051-4bb1-a492-bd1da0a71fcdn%40googlegroups.com.