Contradictions in the effect direction between deltaSVM and QBiC-Pred

[tianwen0003]

Hi I'am using both deltaSVM and QBiC-Pred (http://qbic.genome.duke.edu/makeprediction) to predict the effect of genome variants on TF binding, but for these two method, I always get opposite result for same SNP and same TF. some examples from different TF family are listed below:

• "SNP(hg19)(chr_pos_ref_alt)" "TF" "prediction result from QBiC-Pred" "deltaSVM results"
• "chr10_126278191_C_T" "ETV5" "unbound>bound (mean ETV5 bind allele T other than allele C)" "Loss "
• "chr10_115406307_A_G" "HOXA10" "bound>unbound" "Gain"
• "chr10_35057854_C_A" "ATF3" "bound>unbound" "Gain"

I think there must some problem of that, so I am reporting this issue to both of your groups. I would be very grateful for any reply.

[yjqiu]

@tianwen0003 Thanks for letting us know. We have confidence in our models in general with extensive evaluation as described in the publication. And we cannot comment on other tools since we are not familiar with them.

One thing I want to emphasize is that all predictions are with limitations and users should always use them with caution. Experimental validation should always be the gold standard.

We would keep those examples in mind and may evaluate them in the future. But it is hard to see the truth without further experiments.