pred_perf_r2,pred_perf_pval.pred_perf_qval are always NA

[empathy-c]

I followed the tutorial 'https://github.com/hakyimlab/MetaXcan/wiki/Tutorial:-GTEx-v8-MASH-models-integration-with-a-Coronary-Artery-Disease-GWAS' and it run successfully. About 92 % of model's snps used. But when I check the results table about the spredixcan.py and smultixcan.py ,the column 'pred_perf_r2,pred_perf_pval.pred_perf_qval' only have the results NA in the analysis result of spredixacn.py And the column 't_i_best','t_i_worst' only have the value 'Whole_Blood' or NA in the analysis result of smultixcan.py. Is that correct or there maybe something wrong?
For the two types of analyses, based on single-organism and multi-organism, it seems that I couldn't find any predicted results in the result table. For the single-organism analysis, the column 'pred_perf_r2,pred_perf_pval.pred_perf_qval' are labeled as NA, and the multi-organism 'pred_perf_r2,pred_perf_pval.pred_perf_qval' does not included in the table. As for the analysis of the results, can we consider the results as valid evidence of causal association between the gene and the tissue if the p-value is less than the Bonferroni corrected p-value? I am a beginner in TWAS analysis, and I would greatly appreciate it if you could answer this question for me. Thank you.

[empathy-c]

I then would like to know how to interpret the results obtained from running spredixcan.py and smultixcan.py. What criteria should be used to determine the association between genes and phenotypes?

[ahasuni]

Hi there,

Did you figure out the NA issue? I am running SPrediXcan and am also getting all NAs in pred_perf_r2,pred_perf_pval.pred_perf_qval columns. I am running the program for multiple traits and tissues and its the same result for all of them. Roughly 87% SNPs are included in model.

hasuni

[Fnyasimi]

Hi @empathy-c @ahasuni the pred_perf_r2,pred_perf_pval.pred_perf_qval are metrics of the model when it was trained and not metrics from your S-PrediXcan results. The summary file just gives more information about the model, you should focus on the prediction results with pvalues and zscores