optlang error in optimization

[jotech]

I'm getting an error when trying to compute a condition-specific model by driven.imat. It seems to be an issue with optlang:

AssertionError                            Traceback (most recent call last)
/usr/lib/python3/dist-packages/optlang/expression_parsing.py in _parse_quadratic_expression(expression, expanded)
    128                 term = 1.0 * term
--> 129             assert term.is_Mul, "What is this? {}".format(type(term))
    130             factors = term.args

AssertionError: What is this? <class 'optlang.gurobi_interface.Variable'>

A small example with random expression data to reproduce the problem:

import cobra.test
from numpy.random import random
from numpy import array
from numpy import percentile
import driven

model = cobra.test.create_test_model("textbook")
identifiers = [g.id for g in model.genes]
conditions = ["test"]
expression  = array([random(len(identifiers)) * 100]).transpose()
expression.shape == (len(identifiers), len(conditions))
exprof = driven.ExpressionProfile(identifiers, conditions, expression)

low_cutoff  = percentile(expression, 25)
high_cutoff = percentile(expression, 75)
model.solver = "gurobi" # cplex
model_cond  = driven.imat(model, exprof, (low_cutoff, high_cutoff), condition="test")

Although this error occurs with gurobi and cplex set as solver, it works fine with glpk (in reasonable time only for smaller models).

I tried python 2 and 3. The optlang package is the latest version (1.4.4)

Complete error log

[jotech]

@arabidopsis @Midnighter sorry for pushing this again but I really can't get it run. I tried different python versions (2.7,3.6,3.7), solvers (gurobi,cplex) and systems (debian, centos) and I'm stuck... Could you maybe just have a look if my code above makes sense or do you have another minimal example which is working for you? This would be amazing! I really appreciate any help :)

[Midnighter]

I'm a bit hesitant to step in here because we're not officially supporting this package yet. There's a lot of work to do, we haven't released anything yet, so it's a mixed bag of really old code and some newer code that hasn't been glued together.

[jotech]

hi @Midnighter thanks for your answer! But are there plans to continue this project? Would be really great to have some python code for creating tissue-specific models!

[Midnighter]

Yes, we do want to pick up work on this again. For starters we will focus on diversifying data types, i.e., better support for fluxomics, metabolomics. Even though tissue-specific models are not our speciality, we do want to make transcriptomics and proteomics more useful so there will naturally be some overlap in methods. Timeline-wise you might see something appearing as soon as July.

[Midnighter]

Btw, you should also look around at https://opencobra.github.io/cobrapy/packages/ you might find a fit for your use-case there.