search

drivenbyentropy / aptasuite

A full-featured bioinformatics software collection for the comprehensive analysis of aptamers in HT-SELEX experiments.
https://drivenbyentropy.github.io/
GNU General Public License v3.0
24 stars 11 forks source link

Structure prediction crash #64

Closed PJpb closed 5 years ago

PJpb commented 5 years ago

Hi, The motif analysis tool required structure prediction. I run it (it was rather slow, quite surprising), and it crashed (halted but didn't shut down) was ~20 hours ;) ) [Windows 10 Home]. Here are the logs: AptaSuite log: log_2018-10-20_23-47-06.txt PowerShell log: bug_report_structure_prediction.txt

I'm not sure if this is not an issue of my system rather than AptaSuite, but I can't tell...

edit: HOWEVER! When I restarted the AptaSuite and I tried to run AptaTrace again, this time the software didn't ask for the structures and started to analyze the motifs. I don't know if the database is not corrupted and AptaTrace is able to finish the job - will let know in a few hours. edit 2: As I'd suspected, AptaTrace crushed (the program shut down) when it encountered a corrupted database (but maybe it might be checking the database before starting the analysis?). Here is the log: but_report_aptatrace.txt

drivenbyentropy commented 5 years ago

Hey,

Thanks for reporting this. Let me start with the easier part (your edits): AptaSuite currently does not check explicitly if all the sequences have been predicted before starting with AptaTrace. It only checks if there is a structure database present. I will add explicit checking in the next release. I will also try to modify the prediction stage to continue where it left off, should a crash have occurred. Currently the prediction would have to start from scratch.

As for the actual issue, it seems that the structure prediction did actually complete, but something happened right afterwards when AptaSuite attempted to reopen the database files as read only (to prevent data corruption). I will see if I can replicate this problem here. If you get the chance, could you check if this issue also occurs in a tiny test sample (say 10k unique sequences)?

For the speed issue, it appears your were only using 2 threads for the prediction. If you have more than 2 cores available, you can increase the number of CPUs to use in the Preferences -> Number of Cores menu prior to starting the predictions.

PJpb commented 5 years ago

Hi, It's great to hear that you will add those features, it might save the users some time in the future :)

I wouldn't say that the structure prediction completed - the speed of computation was quite constant, and the ETA was still at about 180h (the calculation of ETA seemed to be ok). I don't think it could speed up overnight and finish the task before I checked it in the morning. I suppose that the software simply froze/halted due to some reason - might be even my system's fault, not the software's...

I was also surprised that only 2 threads out of 8 were used , I thought that this choice is automatic. I'll give it a check.

I'll also try to run a subset of data for the structure prediction, but I won't be able to do it before later this week.

Best, PJ

PJpb commented 5 years ago

Hi, I've checked it with a smaller set of data and it seems to work. The structures got predicted nicely, but it turned out that the dataset is to small for AptaTrace to find any motifs ;) Also, the algorithm did use all the threads, and the computation speed was much higher (but it almost melted down my CPU - this is a different story... ;) ). I'll try to run another analysis with a full selection dataset and see what comes up.

PJpb commented 5 years ago

Hi, I've run structure prediction & AptaTrace with full throttle (which still took over 48 hours ;) ) and it went fine to the very end. So this might have been a random crash after all. I'm closing the thread for now.