Dear HYPHY developpers and contributors,

This is the first time to post a question to this forum.
I'm graduate student in Kurume University in Japan,
and make the best of HYPHY for bacterial genomics.
Thank you for this great software.

This is a question about completely different results
of PARRIS produced by HYPHY ver1.0 and ver2.0.

When I execute HYPHY to conduct PARRIS analysis
by
 $ ./HYPHY my_PARRIS.bf
, output of ver.1.0 was as follows

+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+
|       Model        | Log Likelihood | Synonymous CV |  NS Exp and CV  |  N/S Exp and CV  |  P-Value  | Prm |    AIC    |
+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+
| discr(3), M1a      |    -4918.92369 |    3.18058855 | 0.17065,2.03387 | 1.10154, 2.41569 |    N/A    |   29|    9895.85|
+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+
| discr(3), M2a      |    -4908.81417 |    1.43545198 | 0.49468,2.38712 | 2.08077, 2.75454 | 0.0001529 |   32|    9881.63|

although output of ver.2.0 was as follows

########### 3x3 CLASSES ###########

+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+
|       Model        | Log Likelihood | Synonymous CV |  NS Exp and CV  |  N/S Exp and CV  |  P-Value  | Prm |    AIC    |
+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+
| discr(3), M1a      |    -5111.29881 |    1.39897795 | 0.25228,0.00000 | 1.04456, 0.54217 |    N/A    |   29|   10280.60|
+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+
| discr(3), M2a      |    -5115.76487 |    1.51699935 | 0.22319,0.00000 | 1.56977, 0.70157 | 1.0000000 |   32|   10295.53|
+--------------------+----------------+---------------+-----------------+------------------+-----------+-----+-----------+

Is this a bug in developmental v2.0?
Is the result of v1.0 correct?

The following is the file "my_PARRIS.bf"
which I used for HYPHY v1.0 and v2.0.

inputRedirect = {};
inputRedirect["01"]="Universal";
inputRedirect["02"]="1";
inputRedirect["03"]="/home/k-yahara/HYPHY/test.nex";
inputRedirect["04"]="/home/k-yahara/HYPHY/test.nex.tre";
inputRedirect["05"]="Codon Model";
inputRedirect["06"]="MG";
inputRedirect["07"]="REV";
inputRedirect["08"]="Multi";
inputRedirect["09"]="Default";
inputRedirect["10"]="Dual";
inputRedirect["11"]="Independent";
inputRedirect["12"]="Nucleotide (syn3)";
inputRedirect["13"]="PARRIS";
inputRedirect["14"]="Default";
inputRedirect["15"]="/home/k-yahara/HYPHY/test.nex.parris";
ExecuteAFile("/home/k-yahara/HYPHY/TemplateBatchFiles/PARRIS.bf",  inputRedirect);

Sincerely,

Koji Yahara, Kurume University School of Medicine

Dear Koji,

The result with v2.0 looks strange -- it could be a convergence problem, for sure. M2a is more general than M1a, and the fact that it gets WORSE likelihood points to a software issue.

Would you mind attaching the file so that I could try replicating the problem here and tell you what may be going on?

Sergei

Dear Sergei,

Thank you very much for your kind help.
I attached a compressed file containing "test.nex"
and "test.nex.tre" which were used in "my_PARRIS.bf".

Please uncompress it and execute PARRIS by

$ ./HYPHY my_PARRIS.bf

The content of "my_PARRIS.bf" was shown in my previous post.

Sincerely,

Koji Yahara, Kurume University School of Medicine, Japan

---

Dear Koji,

Thanks so much for bringing my attention to this issue (and an important bug!).

I found the bug in the code for v2.0 (which is only triggered for models with > 2 category variables, like the syn3 model). I just pushed a code update which fixes the source code -- please redownload / recompile HyPhy and confirm that the problem vanishes.

Sergei

Dear Sergei,

Thank you very much!
I recompiled Hyphy and comfirmed that
the problem vanished.

I'll continue to use this great software
in my microbial research.

Sincerely,

Koji Yahara, Kurume University School of Medicine, Japan