Distributional observability of probabilistic Boolean networks
Introduction
Mathematical modeling of biological systems is an attractive way for systematically studying the functions of biological processes and their relationship with the environment. One of the frequently used frameworks for modeling biological systems is that of Boolean models. After assembling individual components and interactions involved in a system, Boolean network (BN) models can properly capture the qualitative temporal behavior of the system without the need for much kinetic details [1]. Typically, a BN is presented in the form of a (deterministic) nonlinear system, while it was shown in [2] that such nonlinear dynamics can always be mapped invertably to a standard discrete-time linear dynamics using an algebraic state representation approach. The main idea behind is that a BN with variables has possible states, and if such states are represented by canonical vectors of length , then the one-step transitions between states can be equivalently specified by a matrix. This formal simplicity makes it comparatively easy to formulate and solve control theoretic problems for BNs, and thereby has stimulated a significant increase in the number of studies in this area [3], [4], [5], [6], [7], [8], [9], [10], [11], [12]. For some recent work on the analysis and control of BNs based on other approaches, see, e.g., [13], [14], [15].
A probabilistic Boolean network (PBN) is a stochastic extension of the classical (deterministic) BN. It may be viewed as a collection of BNs endowed with a probability structure determining the likelihood with which a constituent network is active. PBNs retain the appealing properties of BNs such as requiring few kinetic parameters, but are more robust in the face of uncertainty [16]. The algebraic state representation has also proved a nice framework for addressing control-related problems in PBNs [17], [18], [19], [20], [21], [22], [23].
Observability is one of the main topics in control theory. The classical definition of observability, roughly speaking, says that a system is observable if any two distinct initial states can be distinguished by observing the output of the given system [24]. Observability and related problems for classical BNs have been extensively studied by many authors; see, e.g., [25], [26], [27], [28], and more recently [29], [30], [31], [32], [33], [34], and references therein. Compared with the many studies focusing on BNs, the investigation of the observability of PBNs is currently at an early stage. The authors of [35] considered a type of observability for PBNs, which requires that for any two distinct initial states there is a nonzero probability that different output sequences are produced so that the two states become distinguishable. Such an observability notion was revisited in a recent study [36] and was referred to as finite-time observability in probability. Two slightly different notions of observability were also explored in that study. The first one is termed finite-time observability with probability one, and requires that for any two distinct initial states the corresponding two outputs on a finite time window are distinct with probability one. The second is termed asymptotical observability in distribution and it corresponds to the case where for any pair of distinct initial states, the probability that the outputs are distinct tends to one as the length of the observation window grows to infinity. The authors of [37] suggested an alternative characterization of the property of observability with probability one, and proposed set-theoretic algorithms for determining all possible initial (or final) states compatible with a given output sequence. A necessary and sufficient condition for a multiplex PBN (i.e., a set of PBNs linked through interconnected layers) to be finite-time observable with probability one was given in [38].
We consider in this paper a novel observability problem for PBNs. The problem is to distinguish between the probability distributions of initial states in a PBN from knowledge of the probability distributions of outputs over time. In many existing research works on PBNs, both theoretical and practical, the initial state of the PBN is assumed to be randomly distributed according to a certain probability distribution [39], [40], [41], [42]. This is not only because such an assumption is in better agreement with the stochastic nature of the model, but also due to the fact that there are many real situations in which the initial state of the model cannot be known exactly. We therefore consider initial probability distributions rather than deterministic initial states for starting a PBN (see Remark 1 below for more comments on the motivation). The fact that the initial state is a random variable and that the updating rule of the network is randomly chosen at each time step leads to the fact that the output of a PBN at any given time is also a random variable. Since every stochastic process induces a family of finite dimensional distributions that determines the stochastic process uniquely (see, e.g., [43]), it is then quite reasonable to think that two initial state distributions of a PBN are indistinguishable statistically if the corresponding output processes have the same finite dimensional distributions. Observability based on such indistinguishability is then a direct extension of the classical definition when dealing with PBNs, and is to some extent more stochastic in nature compared to the aforementioned three existing observability notions. Clearly, this type of distributional observability is stronger than the property of finite-time observability in probability, but neither implies nor is implied by the other two observability notions. The main question addressed in this paper is under which conditions a PBN is initial-distribution observable. We provide a complete answer to this question. As in previous studies [35], [36], [37], [38], the theoretical framework developed in this paper is based on the powerful algebraic representation approach.
The remainder of this paper is organized as follows. Section 2 contains the basic notation and briefly reviews PBNs. The main contributions, establishing conditions for distributional observability, are presented in Section 3. Several examples, both contrived and biological, are given in Section 4. A summary of the paper is provided in the final section.
Section snippets
Preliminaries
Let denote the th canonical basis vector, and the set consisting of the canonical vectors . The set of all matrices whose columns are canonical vectors is denoted by . Let and be and matrices, respectively, and let be the least common multiple of and . The (left) semitensor product [44] of and is a matrix, denoted by , and defined by , where is the Kronecker product of matrices and and
Main results
Let us consider a PBN with output, described by the algebraic representation2 As assumed above, is a Bernoulli process with outcome space and probability for the occurrence of outcome , for each , and . We denote by the output stochastic process generated by (6) corresponding to the initial state distribution
Examples
This section presents some examples that illustrate the use of our results.
Example 1 Consider a PBN as in (6), with , , , , , and the selection probabilities given by and . The matrices ’s and ’s are and the subspace is . Direct calculation shows that
Summary
We have addressed in this paper the problem about observability of a PBN. We assume that the initial states of the PBN are not explicitly known but are instead described by probability distributions. The problem is about whether one is able to distinguish between different initial state distributions of the network using only the probability distributions of outputs over time. We have provided a complete solution to this problem, by giving a necessary and sufficient condition for a PBN to be
CRediT authorship contribution statement
Rui Li: Conceptualization, Methodology, Writing – original draft, Writing – review & editing. Qi Zhang: Conceptualization, Writing – original draft, Writing – review & editing. Jianlei Zhang: Methodology, Writing – original draft, Writing – review & editing. Tianguang Chu: Conceptualization, Methodology, Supervision.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
The authors gratefully acknowledge constructive and insightful feedback received from the anonymous reviewers and the Associate Editor.
This research was supported by the National Natural Science Foundation of China under Grants 61503375, 61673027, 62073174, and 62073175, the Humanity and Social Science Youth Foundation of Ministry of Education of China under Grant 20YJCZH228, and the Excellent Young Scholars Funding Project in UIBE under Grant 19YQ10.
References (48)
- et al.
Controllability of Boolean control networks via the Perron-Frobenius theory
Automatica
(2012) - et al.
On algorithms for state feedback stabilization of Boolean control networks
Automatica
(2017) - et al.
Global robust stability and stabilization of Boolean network with disturbances
Automatica
(2017) - et al.
Optimal control of Boolean control networks with average cost: a policy iteration approach
Automatica
(2019) - et al.
On reducible state variables of logical control networks
Systems Control Lett.
(2020) - et al.
Detectability of Boolean networks with disturbance inputs
Systems Control Lett.
(2020) - et al.
Feedback stabilization control design for switched Boolean control networks
Automatica
(2020) - et al.
Minimal controllability of conjunctive Boolean networks is NP-complete
Automatica
(2018) - et al.
State feedback stabilization for probabilistic Boolean networks
Automatica
(2014) - et al.
Sampled-data stabilization of probabilistic Boolean control networks
Systems Control Lett.
(2019)
Controllability of Markovian jump Boolean control networks
Automatica
Logical matrix factorization towards topological structure and stability of probabilistic Boolean networks
Systems Control Lett.
Controllability and observability of Boolean control networks
Automatica
Observability of Boolean networks: a graph-theoretic approach
Automatica
Observability of Boolean networks via set controllability approach
Systems Control Lett.
Redefined observability matrix for Boolean networks and distinguishable partitions of state space
Automatica
Observability of Boolean networks via matrix equations
Automatica
Set reachability and observability of probabilistic Boolean networks
Automatica
Boolean modeling in systems biology: an overview of methodology and applications
Phys. Biol.
A linear representation of dynamics of Boolean networks
IEEE Trans. Automat. Control
On pinning controllability of Boolean control networks
IEEE Trans. Automat. Control
Output tracking of Boolean control networks
IEEE Trans. Automat. Control
A necessary and sufficient graphic condition for the original disturbance decoupling of Boolean networks
IEEE Trans. Automat. Control
Controllability and observability of Boolean networks arising from biology
Chaos
Cited by (7)
Set stabilizability of impulsive probabilistic Boolean networks via impulsive sequence design
2023, Applied Mathematics and ComputationRobust flipping stabilization of Boolean networks: A Q-learning approach
2023, Systems and Control LettersLinear system identifiability from single-cell data
2022, Systems and Control LettersCitation Excerpt :They also use higher moments of the measurement distribution, but restrict to steady-state distributions. A related problem of state observability has been studied by [19] for systems governed by differential equations and by [20] for Boolean models. Assumption (i) could potentially be relaxed, whereas (ii) and (iii) are necessary conditions.
Minimal observability of Boolean control networks
2022, Systems and Control LettersCitation Excerpt :It can multiply two matrices with any dimension and easily transform the logical functions into the algebraic form and has an ability to convert BNs into standard discrete-time linear systems [8]. Based on algebraic form of BNs, many fundamental results have been obtained including observability [9–23], detectability [24,25], synchronization [26–28], controllability [29–33], stabilization [34–37], as well as biological systems [38], etc. Observability is a fundamental concept in system science and control theory.
Asymptotical Controllability of Continuous-Time Probabilistic Logic Control Systems
2024, IEEE Transactions on Automatic ControlFinite-Time Observability of Probabilistic Logical Control Systems
2023, Journal of Systems Science and Complexity