Segmentation fault during inference about igor HOT 6 OPEN

Hi,

I think I solved the first Valgrind error, you were right it was a "typo" leftover from your move to ambiguous nucleotides. I sent you a pull request (which is completely unnecessary for this one-line fix, but I never sent one, and I wanted to try it out)

It didn't solve my issue though, and a few other memory errors are popping out, I'm gonna try to look into these in the next few days (weeks more probably). I'm joining my valgrind error log, if you think your workload is too light,

Best,

Thomas

valgrind_log.txt

from igor.

Hi Thomas,
Thanks for taking the time to produce such a detailed bug report! I highly appreciate it
This indeed seems to be a real segfault issue, valgrind points to the following:

this->dinuc_proba_matrix(i,j) += marginal_array[event_index + i*event_realizations.size() + j];
//TODO This is risky in case the code evolves to have more than 4 realizations

At first I thought I had made a mistake upon making changes to handle ambiguous nucleotides within IGoR but it seems commit 4a1ac62 did not change anything there. I had checked all the code using valgrind at some point but maybe I missed something. Also it seems upon writing this line of code I was already suspicious of its potential pitfalls.

This is indeed the right way to supply a model for an undefined species, however did you make changes to the model_marginals or model_parms by hand or did you use the command line and or cpp interface?

I'll try and find what is wrong just from the error message but I might ask you all the models and sequences if I cannot find an obvious cause.

Best,
Quentin

from igor.

Thanks for the fast answer,

I'm not providing the model_marginals (if I'm reading the documentation correctly, this should initialize it to the uniform distribution), but I write my model_params.txt manually, which could cause the issue.

Here is a summary of the file:

@Event_list
#GeneChoice;V_gene;Undefined_side;7;v_choice
%Omyk_Chr12_NC_035088.1_IGHV20_81397765_81398073;GTTTGCTGTGATGAGATGGATCAGTCTCCTTCTCAGGTGAAAAGACCTGGAGACAAGTTTAAATTGTCATGTAAAATCTCTGGGTTTGATATGACAGACTACTATATGACCTGGATAAGACAGAAACCAGGCAAAGCTCTGGAGTGGATTGGGGGTATTGACTCTGGTTCAACAAATTCTCCAGACTACTCAGACTCCCTGAAAGGCCAGTTCAGCCTGACTGAGGACGTCTCTACAAGCACACAGTTCTTAGAGGCCAAGAGTCTGAGTTCAGAGGACTCTGCTGTTTATTATTGTGCTCGACAGG;0
%Omyk_Chr12_NC_035088.1_IGHV37_81466761_81467069;GTTTGCTGTGATGAGATGGATCAGTCACCTTCTCAGGTGAAAAGACCTGGAGACACGTTTAAATTGTCATGTAAAATCTCTGGGTTTGATATGACAGACTATTTCATGCACTGGATAAGACAGAAACCAGGCAAAGCTCTGGAGTGGATTGGGAGAATTAACTCTGGTTCAAGAAATGCTCCAGTCTACTCAGACTCCCTGAAAGGCCAGTTCACCCTTACTGAGGACATCTCTACAAGCACACAGTTCTTAGAGGCCAAGAGTCTGAGTTCAGAGGACTCTGCTGTTTATTATTGTGCTCGAGAGA;1
%Omyk_Chr12_NC_035088.1_IGHV57_81618381_81618689;GTTTGCTGTGATGAGATGGATCAGTCACCTTCTCAGGTGAAAAGACCTGGAGACACGTTTAAATTGTCATGTAAAATCTCTGGGTTTGATATGACAGACTATTTCATGCACTGGATAAGACAGAAACCAGGCAAAGCTCTGGAGTGGATTGGGGGTATTTACTCTGGTTCAACAGATGCTCCAGTCTACTCAGACTCCCTGAAAGGCCAGTTCATCCTTACTGAGGACATCTCTACAAGCACACAGTTCTTAGAGGCCAAGAGTCTGAGTTCAGAGGACTCTGCTGTTTATTATTGTGCTCGAGAGA;2
#GeneChoice;D_gene;Undefined_side;6;d_gene
%Omyk_Chr12_NC_035088.1_RSS_IGHD1_RSS_81394296_81394994;ACTATATGGGGGC;0
...
%Omyk_Chr13_NC_035089.1_RSS_IGHD11_RSS_48342957_48343024;TATGGGGGCAGC;23
#GeneChoice;J_gene;Undefined_side;7;j_choice
%Omyk_Chr12_NC_035088.1_RSS_JH1_SS_81678720_81678808;ATGCTTTTGACTACTGGGGTAAAGGGACACAAGTCACCGTCTCAACAG;0
...
%Omyk_Chr13_NC_035089.1_RSS_IGHJ10_SS_48368507_48368599;ACAATGCTTTTGACTACTGGGGGAAAGGCACAATGGTTACCGTTTCATCAG;18
#Deletion;V_gene;Three_prime;5;v_3_del
%-4;0
...
%16;20
#Deletion;D_gene;Three_prime;5;d_3_del
%-4;0
...
%16;20
#Deletion;D_gene;Five_prime;5;d_5_del
%-4;0
...
%16;20
#Deletion;J_gene;Five_prime;5;j_5_del
%-4;0
...
%18;22
#Insertion;VD_genes;Undefined_side;4;vd_ins
%0;0
...
%30;30
#Insertion;DJ_gene;Undefined_side;2;dj_ins
%0;0
...
%30;30
#DinucMarkov;VD_genes;Undefined_side;3;vd_dinucl
%T;3
%C;1
%G;2
%A;0
#DinucMarkov;DJ_gene;Undefined_side;1;dj_dinucl
%T;3
%C;1
%G;2
%A;0
@Edges
%GeneChoice_V_gene_Undefined_side_prio7_size3;Deletion_V_gene_Three_prime_prio5_size21
%GeneChoice_V_gene_Undefined_side_prio7_size3;GeneChoice_D_gene_Undefined_side_prio6_size24
%GeneChoice_V_gene_Undefined_side_prio7_size3;GeneChoice_J_gene_Undefined_side_prio7_size19
%GeneChoice_D_gene_Undefined_side_prio6_size24;Deletion_D_gene_Three_prime_prio5_size21
%GeneChoice_D_gene_Undefined_side_prio6_size24;Deletion_D_gene_Five_prime_prio5_size21
%GeneChoice_J_gene_Undefined_side_prio7_size19;Deletion_J_gene_Five_prime_prio5_size23
%GeneChoice_J_gene_Undefined_side_prio7_size19;GeneChoice_D_gene_Undefined_side_prio6_size24
%Deletion_D_gene_Five_prime_prio5_size21;Deletion_D_gene_Three_prime_prio5_size21
@ErrorRate
#SingleErrorRate
0.005

(Only three V-genes, I'm focusing on one primer)

Best,
Thomas

from igor.

Hi Thomas,
Sorry about this I should have spotted it looking more than 2 seconds at the code! Thanks for the PR i'll accept it now!
Keep me updated of your Valgrind quest, I'm afraid because Valgrind have very long runtimes you might not get past the model initialization (maybe I'm wrong, but it will clearly take you some time). Looking at TRA like models could make the computation lighter (although it might hide the issues you are looking for)
Quentin

from igor.

Hi,

I think I figured out how to avoid the segmentation fault, in my model the maximum number of allowed deletions on the D gene was too low , and when I increased it, IGoR ran fine.

So if someone happens to get the same error, increasing the range of parameters of the model seems to work.

While I'm here and because this probably doesn't need a new issue, I have a small question, do you know what happens in IGoR when delD3 + delD5 > len(D) ? (or even delD3 > len(D))

Best,

Thomas

from igor.

Hi Thomas,
Not sure my answer is still relevant but I find this weird that increasing D deletion range would change anything in this segfault thing... I or @alfaceor still need to get a look at this!

As for your last question: the code design assuming there is no deletion simply prevents this from happening. You can only delete up to current D length (meaning if you have added p nucleotides on 1 side, it is still allowed to delete them upon deleting on the other side).

Quentin

from igor.