adelq / dnds Goto Github PK

I'm running dnds with python 2.7 installed via cloning the github repo. When I run the below command

python ~/Dropbox/programs/dnds/dnds.py

It returns the result 0.642189586115. This is different from the result of 0.86 on https://github.com/adelq/dnds

Why is this? Thanks for the nice package!

Sorry for reporting this. I've noticed your code. And it seems that this is not a conventional way of computing dN/dS based on counting. Probably you are not using this code at the moment, but I just want to warn you and people who might use this code.

The classical and improved method of computing dN/dS based on counts are described in the 1986 paper by Nei Gojobori or here. What you are computing looks more like p_N/p_S, which is fine, as long as people are aware that it is another estimate, with different properties. Also, I am not sure that you are correcting for:

1. different trajectories for multinucleotide substitutions (i.e. TTT -> TCT -> TCC and TTT -> TTC -> TCC)
2. stop codons; substitutions involving stop codon should not be considered (for computing N and for mutation trajectories).

There is a reference implementation in C available in PAML (function DistanceMatNG86), it might be a good idea to compare your implementation with it.

Again, sorry for intruding. It's just a friendly message, so you are aware.

Hi - nice to see a python implementation!

Are you able to modify this to specify the genetic code (11 for bacteria) and to take multiple sequence alignments with gaps and Ns and calculate all dnds Vs a reference sequence?

Regards,
Liam

Dear Adel,
I am interested in dn/ds caluclation. Can the result from dnds be compared with yn00 of PAML as the later is vey popular in this aspect?

Best regards,
Hongzhong

The latest version with the dN/dS vs pN/pS split has breaking changes, so it warrants a major version bump.

I have been using this helpful blog post to learn about dN/dS ratios.

http://bioinformatics.cvr.ac.uk/blog/calculating-dnds-for-ngs-datasets/

However, following through their example with your code there was a discrepancy.

According to the scheme set out in that blog post, you are reporting the pN/pS. Following onto the Nei & Gojobori, 1986 paper, which the blog post authors reference, I think the part you omitted was the conversion to per site results according to Jukes & Cantor 1967.

When I add in the below, I get the same result as the blog post authors for their sequence (0.1247)

pn = non_subs / non_sites
ps = syn_subs / syn_sites
dn = -(3 / 4) * log(1 - (4 * pn / 3))
ds = -(3 / 4) * log(1 - (4 * ps / 3))

What are your thoughts on that discrepancy?