I have a list of promoter sequences I extracted using transcription start sites predicted by different computational tools. I want to know how confident are the extracted promoters i.e. how likely are they to be promoter sequences to compare the predictions from the different tools. I am specifically interested in plant promoters. Is there any tool/ method that would help me assess/ predict the confidence of my promoter sequences?
Thank you for your time and help!
This is a very difficult question to answer in general terms.
In my experience, you first need to assess how accurate your TSS predictions are, especially if they are entirely predicted computationally. Which in turn is linked to the quality of the genome assembly and annotation. If TSS is derived from experimental evidence (e.g., CAGE), your results will be considerably better for it.
Promoter predictions will also be linked to promoter class. TATA-box promoters, for example, are relatively easy to detect with most modern tools. However, plant promoters classes are broad, so in silico prediction accuracy will be linked to data quality.
I don't have any recommendation on tools which I realise is what you're asking for, but I thought I would comment on considerations that might be as/more important than tool (I'm sure recent papers/reviews have great suggestions).
One method you could employ, which was common when I last did SV variant calling, is to use multiple callers that use different underlying techniques and accept only ones that overlap to create a conservative but confident dataset.
Thank you! Just to clarify, you mean like an ensemble approach wherein I look if the same TSS is predicted by two or more tools, and then take that to be the confident prediction?
Log in to answer this question.
