How to combine multiple PWMs to identify most likely binding motif

Question

[angemerkel]

I am working with RNA binding proteins trying to identify potential binding sites. I have multiple potential de novo motifs identified from a time series. The data set is very large with multiple replicates, 5 time steps and 2 conditions. The motifs identified from the individual replicates are somewhat inconclusive and difficult to compare. Performing peak calling on pooled replicates acutally gave worse results and using consensus peaks resulted in very few motifs found. I was therefore thinking if it was possible to combine the PWMs instead? Anyone aware of a tool to do so or any other approach to deal with a complex data set (I admit I am quite new to this)? Thanks

[natasha.sernova]

Why do you need multiple PWMs? In my practice some time ago once I had 'double' PWM, but I thought it's an exeption, not a rule. I hope you deal with bacteria, right? Why do you have time series? How did you observe the phenomenon?

[angemerkel]

I am looking for a way to compare many different motifs found in different samples (sets of sequences) with each other, some sort of combining PWM, clustering, etc.The species is human.

[natasha.sernova]

I see, you study a different problem. in my opinion PWM is 'position weight matrix' "How does a position weight matrix work? Definition. A position weight matrix (PWM) is a commonly used representation of motifs (patterns) in biological sequences (Ben-Gal et al. 2005). In a position weight matrix, each row corresponds to one symbol of the alphabet, e.g., amino acids or nucleic acids, and each column corresponds to one position in the pattern". It works well for bacterial TransFactors. I am not sure about humans. Good luck!