Hi all,
I have a dataset from a base editing experiment where some spacers show efficient editing, while others show little or no editing. In addition, among the edited sites, some show the intended conversion (e.g. A to G), whereas others show unintended or “mis-editing” events (e.g. A to C or A to T).
I would like to identify sequence patterns or motifs in the flanking regions that may explain:
- Why certain target bases are more prone to editing than others
- Why some targets undergo correct editing versus incorrect base
conversions
Specifically, I am interested in approaches to analyze the local sequence context (e.g. 1-3 bp around the edited base) to discover motifs associated with:
- High vs low editing efficiency
- Correct (A to G) vs incorrect edits
Has anyone worked on similar analyses, or can recommend computational/statistical methods or tools (e.g. motif discovery, k-mer enrichment, ML approaches) that are suitable for this type of problem?
Any suggestions or references would be greatly appreciated.
1 answer
Worth checking BE-Hive first (Arbab et al. 2020) - trained on tens of thousands of target sites and predicts both efficiency and the outcome spectrum, so you may find it already explains a decent chunk of your variance before you build anything.
The main thing I'd watch in your own analysis is that position within the editing window will swamp sequence context. An A at protospacer position 6 edits far better than one at position 10 regardless of what sits next to it, so if your efficient and inefficient spacers differ in where the target A falls, whatever "motif" you find is really just position. Stratify by position first, then look at flanking base within each. The known ABE context effect is a single-position one anyway - the base immediately 5', roughly TA > AA > CA ~ GA - which is also why MEME or HOMER aren't the right instrument here, they're built for longer motifs in unaligned sets.
The A to C and A to T calls I'd treat with suspicion until you've checked the alignments. Misalignment near indels produces exactly that pattern at edited positions.
Log in to answer this question.