Any reason for higher reported levels of non-CG methylation when analysing non-directional WGBS libraries with Bismark?

Question

[benbio]

Hi, just FYI I've also posted a variation of this question on seqanswers, I'll update both posts accordingly with any responses.

---

I've been analysing some WGBS libraries for a couple of different insect species, using Bismark for alignment. For context in most insects methylation levels are low to begin with and non-CG methylation is often considered to be noise.

I have both directional and non-directional libraries for each species, with alignment performed using the non-directional parameter in Bismark for the latter. I've noticed a pattern of non-directional libraries having higher levels of non-CG methylation (as estimated by Bismark's splitting report generated during methylation extraction). Examples as follows from this report - same species, both control groups:

Directional:

Final Cytosine Methylation Report
=================================
Total number of C's analysed:   1441325271

Total methylated C's in CpG context:    10947321
Total methylated C's in CHG context:    1518883
Total methylated C's in CHH context:    7907753

Total C to T conversions in CpG context:    260501288
Total C to T conversions in CHG context:    203098923
Total C to T conversions in CHH context:    957351103

C methylated in CpG context:    4.0%
C methylated in CHG context:    0.7%
C methylated in CHH context:    0.8%

Non-directional:

Final Cytosine Methylation Report
=================================
Total number of C's analysed:   1671579979

Total methylated C's in CpG context:    15071917
Total methylated C's in CHG context:    3393242
Total methylated C's in CHH context:    17650398

Total C to T conversions in CpG context:    360463445
Total C to T conversions in CHG context:    264082345
Total C to T conversions in CHH context:    1010918632

C methylated in CpG context:    4.0%
C methylated in CHG context:    1.3%
C methylated in CHH context:    1.7%

This pattern holds in all samples and in both species I'm studying. Before I conclude that this is the result of quirks of these specific datasets, I was wondering if anyone was aware of a reason to do with the nature of non-directional libraries as to why the non-CG methylation might be reported at a higher level?

[mark.ziemann]

It could be:

1. Difference in bisulfite conversion efficiency
2. Overall error rate differences in these seq runs - if you could obtain the phi-x from those runs you could check the error rates precisely
3. Something related to the alignment search space being twice as large for non-directional libraries. Perhaps try to replicate the result with another aligner like Biscuit

[benbio]

Thanks for your suggestions, Mark. Point 3 is the kind of thing I was thinking might be the reason, as CpG context is unaffected, reported at similar levels in both directional and non-directional libraries - I assume such issues as 1 and 2 would affect all contexts. I'll try your suggestion of using alternative aligners - thanks again.