`DeepPrimeGuideRNA`: reverse-complement of all sequences in input does not always give same output?

[francoiskroll]

Sorry, I think I'm almost done with all my sanity checks 😃. Just want to make sure I'm using the tool correctly before doing anything substantial.

With DeepPrimeGuideRNA(...), why does it matter—sometimes, but not always—whether one works on the 5'–3' genome strand or 3'–5' genome strand?

If I take the example from documentation:

## target in 5'–3' genome direction ##
# PBS & RT on 3'–5' strand
# i.e. exactly the example in documentation
target='ATAAAAGACAACACCCTTGCCTTGTGGAGTTTTCAAAGCTCCCAGAAACTGAGAAGAACTATAACCTGCAAATG'
pbs='GGCAAGGGTGT'
rtt='CGTCTCAGTTTCTGGGAGCTTTGAAAACTCCACAA'
edit_len=1
edit_pos=34
edit_type='sub'
peg = DeepPrimeGuideRNA('pegRNA_test', target=target, pbs=pbs, rtt=rtt, 
                           edit_len=edit_len, edit_pos=edit_pos, edit_type=edit_type)
score = peg.predict('PE2')
print(score) # 0.09

## reverse-complement of all three sequences, so target is now in 3'–5' genome direction ##
target='CATTTGCAGGTTATAGTTCTTCTCAGTTTCTGGGAGCTTTGAAAACTCCACAAGGCAAGGGTGTTGTCTTTTAT'
pbs='ACACCCTTGCC'
rtt='TTGTGGAGTTTTCAAAGCTCCCAGAAACTGAGACG'
edit_len=1
edit_pos=34
edit_type='sub'
peg = DeepPrimeGuideRNA('pegRNA_test', target=target, pbs=pbs, rtt=rtt, 
                           edit_len=edit_len, edit_pos=edit_pos, edit_type=edit_type)
score = peg.predict('PE2')
print(score) # 0.16

Another example:

## target in 3'–5' genome direction ##
# because gene (zebrafish adgrf3b) is in reverse direction
# PBS & RT on 5'–3' strand
target='CAAATGTGTATGGCAGATGTCCAGAGGCGACGAAGCGCCATTGACTTTGGGACCAGGGAGTGAAATAATGCTTT'
pbs='GACATCTGCC'
rtt='GCTTCGTCGCCTCTG'
edit_len  = 1
edit_pos  = 5
edit_type = 'sub'
peg = DeepPrimeGuideRNA('pegRNA', target=target, pbs=pbs, rtt=rtt, 
                           edit_len=edit_len, edit_pos=edit_pos, edit_type=edit_type)
score=peg.predict('PE2')
print(score) # 0.54

## reverse-complement of all three sequences, so target is now in 3'–5' genome direction ##
target='AAAGCATTATTTCACTCCCTGGTCCCAAAGTCAATGGCGCTTCGTCGCCTCTGGACATCTGCCATACACATTTG'
pbs='GGCAGATGTC'
rtt='CAGAGGCGACGAAGC'
edit_len  = 1
edit_pos  = 5
edit_type = 'sub'
peg = DeepPrimeGuideRNA('pegRNA', target=target, pbs=pbs, rtt=rtt, 
                           edit_len=edit_len, edit_pos=edit_pos, edit_type=edit_type)
score=peg.predict('PE2')
print(score) # 1.24

I would have thought reverse-complementing all sequences in input would always give the same output. Am I missing something obvious?

---

EDIT: Corrected first example, as per reply below from @hkimlab.

[Goosang-Yu]

I think some codes should be corrected.

In this part,

peg = DeepPrimeGuideRNA('pegRNA_test', target=target, pbs=pbs, rtt=rtt, 
                           edit_len=edit_len, edit_pos=edit_pos, edit_type=edit_type)

score = pegrna.predict('PE2')

score should be as below:

score = peg.predict('PE2')

Maybe pegrna instance already have some data about DeepPrime pipeline

---

Also, DeepPrimeGuideRNA dose not support reverse-complemented sequence form.
So, even if it returns some score, it's not actual pegRNA activity score when it was given as input of proper direction and position.

ToDo It seems DeepPrimeGuideRNA should have input format checker module.

[francoiskroll]

Ouch, my bad, thanks for noticing. I'll edit in my question so it's not misleading. In both examples, reverse-complementing all three sequences does not give the same output then.

In my example, can you confirm which one is correct? I'm getting myself confused when the gene is reverse. Do we always want target to be 5'–3' genome?

Thanks!

EDIT – Actually, I think my confusion is about the way DeepPrime handles the PBS & RT sequence, possibly arising from #88 where it looked liked webtool & genET (DeepPrime Python package) could give sPBS_RTSeq in opposite directions for the same pegRNA. Do we want to give the pegRNA's PBS & RT as the actual RNA sequence (just Us replaced by Ts) or reverse-complement of that? From the documentation example, I would say RNA sequence?

[Goosang-Yu]

The input format for DeepPrimeGuideRNA should be like below:

So, for your gene of interest, the first one is correct:

## target in 3'–5' genome direction ##
# because gene (zebrafish adgrf3b) is in reverse direction
# PBS & RT on 5'–3' strand
target='CAAATGTGTATGGCAGATGTCCAGAGGCGACGAAGCGCCATTGACTTTGGGACCAGGGAGTGAAATAATGCTTT'
pbs='GACATCTGCC'
rtt='GCTTCGTCGCCTCTG'
edit_len  = 1
edit_pos  = 5
edit_type = 'sub'
peg = DeepPrimeGuideRNA('pegRNA', target=target, pbs=pbs, rtt=rtt, 
                           edit_len=edit_len, edit_pos=edit_pos, edit_type=edit_type)
score=peg.predict('PE2')
print(score) # 0.54

Please see this annotated sequence map also:

However, the RTT sequence dose not have any prime editing information. This sequence looks exactly same to part of target sequence. rtt sequence should contain mismatch for inducing desired prime editing.

[Goosang-Yu]

I've added additional explanations and illustrations to the GenET documentation regarding this issue. I hope it helps when you're using it.

If there are any parts that are not clear, feel free to let me know anytime.

PS. Also, in your example (zebrafish gene), it seems like there is no intended prime edit in the RT template.