Inconsistent tss_distance in cis.map_nominal and cis.map_independent

[anglixue]

Inconsistent tss_distance in cis.map_nominal and cis.map_independent

[francois-a]

Hi, I'm not able to reproduce this. Can you please check if this still occurs with v1.0.7?

[anglixue]

Hi Francois,

Thanks for your reply.

Just upgraded to v1.0.7 but encountered an error when importing the mixqtl.py from the tensorqtl. It seems to be a minor import syntax error in line#59 of mixqtl.py induced by a typo I believe.

I've tested it on my machine and after removing the asterisk it's good to run. I've submitted a pull request for you to review.

Unfortunately, the inconsistency of tss_distance between cis.map_nominal and indep_df was still there.

[anglixue]

Hi @francois-a ,

I have used v1.0.7 and re-ran all the analyses. I found that there are always ~10% variants that showed inconsistent results between cis.map_nominal() and cis_df().

Here is the example for chromosome 22 in my data. I tested 408 genes and 69805 variants. After matching nominal associations with significant associations, 43 / 408 eQTLs showed inconsistent pval_nominal, slope/slope_se, tss_distance. The allele freq, ma_samples, ma_count.

Nominal associations:

head(nom[index, -1])
    phenotype_id  variant_id tss_distance         af ma_samples ma_count pval_nominal         slope     slope_se                   pair
102         DDTL 22:24266726       -42364 0.38367346        603      752 3.084511e-34  0.0054795183 0.0004317051       22:24266726_DDTL
103   KB-226F1.2 22:24267047       -44354 0.38265306        602      750 4.183513e-07  0.0009024401 0.0001770988 22:24267047_KB-226F1.2
104          DDT 22:24245292       -68263 0.33214286        525      651 2.866474e-19 -0.0525704920 0.0057343070        22:24245292_DDT
136         HSCB 22:28339553      -798467 0.15306123        263      300 7.034724e-01  0.0009122025 0.0023957987       22:28339553_HSCB
137      CCDC117 22:28303075      -865588 0.05408163        104      106 1.195958e-01 -0.0037228062 0.0023896908    22:28303075_CCDC117
138         XBP1 22:28629713      -560831 0.16938776        296      332 1.304909e-30  0.1073389800 0.0090110050       22:28629713_XBP1

Significant associations:

head(sig[index,colnames(nom)[-1]])
    phenotype_id  variant_id tss_distance         af ma_samples ma_count pval_nominal        slope     slope_se                   pair
102         DDTL 22:24266726       -46829 0.38367346        603      752 9.790065e-49 -0.082714660 0.0053232773       22:24266726_DDTL
103   KB-226F1.2 22:24267047       -42043 0.38265306        602      750 2.517564e-34  0.005490042 0.0004319025 22:24267047_KB-226F1.2
104          DDT 22:24245292       -66109 0.33214286        525      651 1.776509e-08  0.001006383 0.0001771545        22:24245292_DDT
136         HSCB 22:28339553      -940028 0.15306123        263      300 3.107622e-04  0.001176659 0.0003251019       22:28339553_HSCB
137      CCDC117 22:28303075      -834945 0.05408163        104      106 2.647748e-03  0.012170344 0.0040382070    22:28303075_CCDC117
138         XBP1 22:28629713      -538950 0.16938776        296      332 1.030225e-04  0.005427211 0.0013917054       22:28629713_XBP1

And if I plot the pval_nominal between the two results, it looks like this

The allele frequency is consistent

Hope to get some advice from you on how I can dig into and solve the problem.

Thank you!

Cheers,
Angli

[francois-a]

Hi, I'm unable to reproduce this. The TSS distances between the different modes (cis permutations, nominal, independent) are consistent in my tests. Do you have duplicate gene symbols in your inputs? Otherwise, can you share a small subset of your dataset that reproduces the problem?

Edit: resolved (not a TensorQTL issue).