* Package name    : qmean
  Version         : 4.2.0
  Upstream Author : Gabriel Studer, Pascal Benkert and Marco Biasini
* URL             : https://github.com/lindenb/makefile2graph
* License         : Apache-2.0
  Programming Lang: C++, Python
  Description     : Qualitative Model Energy ANalysis

QMEAN is used to assess model quality of constructed protein models.
Having the ability to run QMEAN locally in command line environment
would complete the pipeline of protein model building in Debian.

QMEAN requires libdevel packages of OpenStructure, which I have just
uploaded.

QMEAN relies on a pre-trained neural network (most likely keras-based),
which is shipped with the package. According to Debian Machine Learning
Policy [1], pre-trained neural networks must be reproducible, thus
upstream has to be contacted for the details to train the network.

[1] https://salsa.debian.org/deeplearning-team/ml-policy/-/tags/v0.2.5

Hello,

I have pushed my packaging efforts to salsa [1]. I hope to maintain
qmean under DebiChem Team umbrella.

Currently, a couple tests fail. Pre-trained models are not excluded yet.

[1] https://salsa.debian.org/debichem-team/qmean

Andrius

Hi,

This is sad news for qmean. I got the package up and running and I use
it quite a lot (built locally currently due to unresolved issues with
pretrained network weights). The only viable solution I see right now is
to ask qmean's upstream to migrate away from keras.

Best,
Andrius

fall [1], again with support for multiple backends including less
problematic pytorch. Maybe it is worth to keep src:keras for the time
being, until keras 3 arrives?

[1] https://keras.io/keras_core/announcement/

Andrius