Are PWMs the default way of representing motifs?

Question

[dutta.ritik]

Are PWMs (Position Weight Matrices) the most accepted way of representing motifs? Are they effective or are there any drawbacks to using PWMs? Are there any alternatives to using PWMs to represent motifs?

Answer 1

[i.sudbery]

PWMs are almost certainly the most common way of representing a TF motif. There are variants that give a frequency, or a count rather than a weight, but they are effectively the same.

The main drawback of PWMs is that they treat each position as independent. That is they can't encode the situation where a motif can be ATTC or AGGC, but not ATGC or AGTC. There is some debate about how common such a situation is, or how much it matters. There have been some attempts to use HMMs to represent motifs, but they never really caught on, mostly do to the complexity and difficulty in converting the parameters in a HMM into something a human can intuitively understand as a motif.

[dutta.ritik]

Thank you for your reply. I am majoring in Computer Science but am working on a project involving motif discovery, and using Finite State Machines/ HMMs made more sense to me since they encode dependence on the previously chosen nucleotides. Are you aware of any papers that use HMMs, or discuss the drawbacks/advantages of different methods for representing motifs? My initial search on HMMs led me to this paper from 2009: https://www.ncbi.nlm.nih.gov/pubmed/19210737, though I haven't read it yet. Thanks.