initialConcentration ignored (species initialized to 0) when its compartment is set by an assignment rule

[wshlavacek]

If a species has an initialConcentration and lives in a compartment whose size is governed by an assignmentRule,
roadrunner initializes the species concentration to 0 instead of the stated value. The compartment size itself
comes out right. Only the species is wrong.

Here's a minimal case. X should start at 0.05 and the compartment V starts at 1, so nothing about the rule
changes the value at t=0:

import roadrunner
 
 sbml = """<?xml version="1.0" encoding="UTF-8"?>
 <sbml xmlns="http://www.sbml.org/sbml/level3/version1/core" level="3" version="1">
   <model id="init_bug">
     <listOfCompartments>
       <compartment id="V" spatialDimensions="3" size="1" constant="false"/>
     </listOfCompartments>
     <listOfSpecies>
       <species id="X" compartment="V" initialConcentration="0.05"
                hasOnlySubstanceUnits="false" boundaryCondition="false" constant="false"/>
     </listOfSpecies>
     <listOfRules>
       <assignmentRule variable="V">
         <math xmlns="http://www.w3.org/1998/Math/MathML">
           <apply><plus/>
             <cn type="integer">1</cn>
             <csymbol encoding="text" definitionURL="http://www.sbml.org/sbml/symbols/time">time</csymbol>
           </apply>
         </math>
       </assignmentRule>
     </listOfRules>
   </model>
 </sbml>"""

rr = roadrunner.RoadRunner(sbml)
print(rr["[X]"], rr["V"]) # prints 0.0 1.0 ; expected 0.05 1.0

Expected [X] = 0.05. Actual is 0.0 (roadrunner 2.9.2).

A few things I narrowed down:

• It only happens when the compartment rule is time-dependent. Make the compartment constant, or give it an
  assignmentRule with a constant value (which gets folded), and X comes out 0.05 as expected.
• The wrong value depends on the rule expression. V := 1 + time gives 0; V := 2/(1 + time) gives -inf; V := 2 +
  0*time gives nan. So it looks like the species amount is being computed against a compartment volume that hasn't
  been evaluated yet, rather than the initialConcentration being used directly.
• The compartment itself reads back correctly at t=0, so it's specifically the species initialization that's off.

I ran into this on the BioModels curated model BIOMD0000000342 (Zi 2012 TGF-beta), where the medium compartment Vmed
has an assignmentRule and the extracellular ligand TGF_beta_ex has initialConcentration 0.05. roadrunner loads
it as 0, which leaves the whole receptor cascade unstimulated.