In these days I was performing TF binding sites prediction using FIMO (MEME Suite). I am trying to check if there are TF binding sites for TF related to biological processes at the promoter level of the gene in my interest.
I am performing the analysis using;
- Proximal promoter sequence (-800 / +200 from TSS)
- promoter sequence region annotated in the ENSEMBL regulatory feature (-1600 / +1800 from TSS)
- JASPAR 2018 PWM
In some cases I have obtained results with very high q values (like 0,50; but this was not always the case) and, in general, I know that this kind of predictions may result in many false positives. However, by contextualizing the position of the predicted binding sites (and others that are known to be near to each other) I got results that make sense.
I was wondering if there are other computational methods for increasing prediction reliability. By trying to visualize ChIP-seq experiment peaks I have seen that there were correspondences in predictions regions and peaks. The same thing for checking identity with the promoter of the orthologous gene in mouse. However, I have to admit that my approach in doing that was very naive and I am not sure it is correct.
Thank you for your support! Daniele