StructuralIdentifiability.jl

Fast and automatic structural identifiability software for ODE systems
Popularity
110 Stars
Updated Last
25 Days Ago
Started In
August 2020

StructuralIdentifiability.jl

Join the chat at https://julialang.zulipchat.com #sciml-bridged Global Docs

codecov Build Status

ColPrac: Contributor's Guide on Collaborative Practices for Community Packages SciML Code Style

About

StructuralIdentifiability.jl is a Julia package for assessing structural parameter identifiability of parametric ODE models, both local and global. This includes computation of identifiable functions of states and parameters. The package also offers functionality to assess local identifiability in discrete-time models. For an introduction to structural identifiability, we refer to [2].

How to install

The package can be installed from this repository by

using Pkg
Pkg.add("StructuralIdentifiability")

Tutorials and Documentation

For information on using the package, see the stable documentation. Use the in-development documentation for the version of the documentation, which contains the unreleased features.

How to use

The package can be loaded by using StructuralIdentifiability.

Defining a system

A parametric ODE system in the state-space from can be defined by the @ODEmodel macro:

ode = @ODEmodel(
    x1'(t) = -(a01 + a21) * x1(t) + a12 * x2(t) + u(t),
    x2'(t) = a21 * x1(t) - a12 * x2(t) - x3(t) / b,
    x3'(t) = x3(t),
    y(t) = x2(t)
)

In this example:

  • x1(t), x2(t), x3(t) are the state variables, they defined the state of the system and are assumed to be unknown;
  • u(t) is the input/control variable which is assumed to be known and generic (exciting) enough;
  • y(t) is the output variable which is assumed to be observed in the experiments and, thus, known;
  • a01, a21, a12, b are unknown scalar parameters.

Note that there may be multiple inputs and outputs.

Assessing identifiability

The identifiability of the parameters in the model can be assessed by the assess_identifiability function as follows

assess_identifiability(ode)

The returned value is a dictionary from the parameter of the model to one of the symbols

  • :globally meaning that the parameter is globally identifiable
  • :locally meaning that the parameter is locally but not globally identifiable
  • :nonidentifiable meaning that the parameter is not identifiable even locally.

For example, for the ode defined above, it will be

OrderedDict{Any, Symbol} with 7 entries:
  x1(t) => :locally
  x2(t) => :globally
  x3(t) => :nonidentifiable
  a01   => :locally
  a12   => :locally
  a21   => :globally
  b     => :nonidentifiable

If one is interested in the identifiability of particular functions of the parameter, one can pass a list of them as a second argument:

assess_identifiability(ode, funcs_to_check = [a01 + a12, a01 * a12])

This will return:

OrderedDict{Any, Symbol} with 2 entries:
  a01 + a12 => :globally
  a01*a12   => :globally

Finding identifiable functions

In the example above, we saw that, while some parameters may be not globally identifiable, appropriate functions of them (such as a01 + a12 and a01 * a12) can be still identifiable. However, it may be not so easy to guess these functions (even in this example). Good news is that this is not needed! Function find_identifiable_functions can find generators of all identifiable functions of a given model. For instance:

find_identifiable_functions(ode)

will return

3-element Vector{AbstractAlgebra.Generic.Frac{Nemo.QQMPolyRingElem}}:
 a21
 a01*a12
 a01 + a12

which are exactly the identifiable functions we have found before. Furthermore, by specifying with_states = true, one can compute the generating set for all identifiable functions of parameters and states (in other words, all observable functions):

find_identifiable_functions(ode, with_states = true)

This will return

6-element Vector{AbstractAlgebra.Generic.Frac{Nemo.QQMPolyRingElem}}:
 x2(t)
 a21
 a01*a12
 a01 + a12
 x3(t)//(a12*b + a21*b + b)
 (-x1(t)*a21*b + x2(t)*a12*b + x2(t)*a21*b + x2(t)*b + x3(t))//(a21*b)

Assessing local identifiability

Local identifiability can be assessed efficiently even for the models for which global identifiability analysis is out of reach. This can be done using the assess_local_identifiability function, for example:

assess_local_identifiability(ode)

The returned value is a dictionary from parameters and state variables to 1 (is locally identifiable/observable) and 0 (not identifiable/observable) values. In our example:

OrderedDict{Any, Bool} with 7 entries:
  x1(t) => 1
  x2(t) => 1
  x3(t) => 0
  a01   => 1
  a12   => 1
  a21   => 1
  b     => 0

As for assess_identifiability, one can assess local identifiability of arbitrary rational functions in the parameters (and also states) by providing a list of such functions as the second argument.

Remark The algorithms we used are randomized, the default probability of the correctness of the result is 99%, one can change it by changing the value of a keyword argument p to any real number between 0 and 1, for example:

# pobability of correctness 99.9%
assess_identifiability(ode; p = 0.999)

Contacts

Maintained by Gleb Pogudin (gleb.pogudin@polytechnique.edu)

References

[1] Ruiwen Dong, Christian Goodbrake, Heather Harrington, and Gleb Pogudin, Differential elimination for dynamical models via projections with applications to structural identifiability, preprint, 2021.

[2] Hongyu Miao, Xiaohua Xia, Alan S. Perelson, and Hulin Wu, On Identifiability of Nonlinear ODE Models and Applications in Viral Dynamics, SIAM Review, 2011.

[3] Alexey Ovchinnikov, Anand Pillay, Gleb Pogudin, and Thomas Scanlon, Computing all identifiable functions for ODE models, preprint, 2020.

[4] Alexandre Sedoglavic, A probabilistic algorithm to test local algebraic observability in polynomial time, Journal of Symbolic Computation, 2002.

Used By Packages