## From Boolean to Probabilistic Boolean Networks as Models of Genetic Regulatory Networks (2002)

Venue: | Proc. IEEE |

Citations: | 91 - 17 self |

### BibTeX

@INPROCEEDINGS{Shmulevich02fromboolean,

author = {Ilya Shmulevich and Edward R. Dougherty and Wei Zhang},

title = {From Boolean to Probabilistic Boolean Networks as Models of Genetic Regulatory Networks},

booktitle = {Proc. IEEE},

year = {2002},

pages = {1778--1792}

}

### Years of Citing Articles

### OpenURL

### Abstract

Mathematical and computational modeling of genetic regulatory networks promises to uncover the fundamental principles governing biological systems in an integrarive and holistic manner. It also paves the way toward the development of systematic approaches for effective therapeutic intervention in disease. The central theme in this paper is the Boolean formalism as a building block for modeling complex, large-scale, and dynamical networks of genetic interactions. We discuss the goals of modeling genetic networks as well as the data requirements. The Boolean formalism is justified from several points of view. We then introduce Boolean networks and discuss their relationships to nonlinear digital filters. The role of Boolean networks in understanding cell differentiation and cellular functional states is discussed. The inference of Boolean networks from real gene expression data is considered from the viewpoints of computational learning theory and nonlinear signal processing, touching on computational complexity of learning and robustness. Then, a discussion of the need to handle uncertainty in a probabilistic framework is presented, leading to an introduction of probabilistic Boolean networks and their relationships to Markov chains. Methods for quantifying the influence of genes on other genes are presented. The general question of the potential effect of individual genes on the global dynamical network behavior is considered using stochastic perturbation analysis. This discussion then leads into the problem of target identification for therapeutic intervention via the development of several computational tools based on first-passage times in Markov chains. Examples from biology are presented throughout the paper. 1

### Citations

8313 |
Genetic Algorithms
- Goldberg
- 1989
(Show Context)
Citation Context ...idea was already considered by Kauffman and Levin [92, 20] and corresponds to the bit-flipping mutation operator in NK Landscapes. This also corresponds to the mutation operator in genetic algorithms =-=[93]. Th-=-is type of ‘randomization’, namely allowing genes to randomly flip value, is biologically meaningful. Since the genome is not a closed system, but rather has inputs from the outside, it is known t... |

7493 |
Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference
- Pearl
- 1988
(Show Context)
Citation Context ...terministic) predictor under uncertainty. Although reasoning under uncertainty is not a new problem and has been extensively studied in the artificial intelligence and pattern recognition communities =-=[85, 86]-=-, it nonetheless presents a problem when the uncertainty cannot be reliably estimated. Without doing so, we cannot know how well the designed predictor generalizes over the population. In other words,... |

1754 | A theory of the learnable
- Valiant
- 1984
(Show Context)
Citation Context ...tworks has focused on identifying the network structure from gene expression data [58, 59, 60, 61, 62, 63, 64, 65, 66]. At the same time, a large body of related work in computational learning theory =-=[10, 67]-=- has addressed very similar problems, namely, learning or inferring Boolean functions from examples of their input-output behavior 5 . A major focus in this field has been on the construction of algor... |

1487 |
System Identification: Theory for the User
- Ljung
- 1987
(Show Context)
Citation Context ...ver and understand the underlying gene regulatory mechanisms by means of inferring them from data. This generally falls within the scope of computational learning theory [10] or system identification =-=[11]-=-. Second, by using the inferred model, we endeavor to make useful predictions by mathematical analysis and computer simulations. There is a natural order to these two objectives in that the inference ... |

1356 | Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring
- Golub
- 1999
(Show Context)
Citation Context ...tial of gene expression data has already been observed. For example, cancer classification using a variety of methods has been used to exploit the class-separating power of expression data: leukemias =-=[6]-=-, various cancers [7], small, round, blue-cell cancers [8], and hereditary breast cancer [9]. The next step is to dig deeper and understand the underlying mechanisms and the functions of genes in heal... |

1236 |
Image analysis and mathematical morphology
- Serra
- 1982
(Show Context)
Citation Context ...ch as median filters, stack filters, and morphological filters [37]–[42]. A key instance of root signals occurs in the case of idempotent filters, which play a central role in mathematical morpholog=-=y [43]-=-, [44]. Consider a binary-valued one-dimensional (1-D) signal of arbitrary length. Suppose a window of length is sliding across this signal. At every location of the window, the contents inside the wi... |

1215 |
The origins of order: self-organization and selection in evolution
- Kauffman
- 1993
(Show Context)
Citation Context ... most attention, not only from the biology community, but also in physics, is the Boolean Network model, originally introduced by Kauffman [16, 17, 19, 18]. Good reviews of this model can be found in =-=[20, 21, 22]-=-. In this model, gene expression is quantized to only two levels: ON and OFF. The expression level (state) of each gene is functionally related to the expression states of some other genes, using logi... |

1094 |
Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science 270:467–470
- Schena, Shalon, et al.
- 1995
(Show Context)
Citation Context ...development of high-throughput genomic and proteomic technologies is empowering researchers in the collection of broad-scope gene information. The advent of cDNA microarrays and oligonucleotide chips =-=[1]��-=-�[5], which facilitate large-scale surveys of gene expression, has incited much interdisciplinary scientific activity. The diagnostic potential of gene expression data has already been observed. For e... |

907 |
Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization,” Molecular Biology of the Cell
- Spellman, Sherlock, et al.
- 1998
(Show Context)
Citation Context ... for cell cycle regulation. For example, recent studies with microarray technology have revealed a comprehensive catalog of yeast genes whose transcript levels vary periodically within the cell cycle =-=[55]-=-. Another interpretation of the attractors in Boolean networks is that they represent cellular states, such as proliferation (cell cycle), apoptosis (programmed cell death), and differentiation (execu... |

660 |
J.D.Watson. Molecular Biology of the Cell
- Alberts, Bray, et al.
- 1989
(Show Context)
Citation Context ...nt estimate for the number of genes in the human genome is almost 40,000, this would imply that there are roughly 200 different cellular types. It is known that adult humans have about 254 cell types =-=[53]. -=-It was also shown (see [20]) that the expected cycle length for a K =2Boolean network is also on the order of √ n. The cyclical nature of the attractors can be equated to the mitotic cycle in cells.... |

477 | Modeling and simulation of genetic regulatory systems: A literature review
- Jong
- 2002
(Show Context)
Citation Context ...models, logical networks, Petri nets, graph-based models, grammars, and process algebras. There have been a number of excellent survey papers on modeling and simulation of genetic regulatory networks =-=[12, 13, 14]-=- as well as a recent book [15]. 2 Why Boolean? The model system that has received, perhaps, the most attention, not only from the biology community, but also in physics, is the Boolean Network model, ... |

409 | Matabolic stability and epigenesis in randomly constructed genetic nets
- Kauffman
- 1969
(Show Context)
Citation Context ...Why Boolean? The model system that has received, perhaps, the most attention, not only from the biology community, but also in physics, is the Boolean Network model, originally introduced by Kauffman =-=[16, 17, 19, 18]-=-. Good reviews of this model can be found in [20, 21, 22]. In this model, gene expression is quantized to only two levels: ON and OFF. The expression level (state) of each gene is functionally related... |

288 | Ratio-based decisions and the quantitative analysis of cDNA microarrays images
- CHEN, DOUGHERTY, et al.
- 1997
(Show Context)
Citation Context ...when it is binarized. This leads to the second question. In the case of cDNA microarray data, it is widely recognized that reproducibility of measurements and between-slide variation is a major issue =-=[28, 29]. Fu-=-rthermore, genetic regulation exhibits considerable uncertainty on the biological level. Indeed, evidence suggests 3sthat this type of “noise” is in fact advantageous in some regulatory mechanisms... |

286 |
Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks. Nat Med 7:673–679
- Khan, JS, et al.
- 2001
(Show Context)
Citation Context ...r example, cancer classification using a variety of methods has been used to exploit the class-separating power of expression data: leukemias [6], various cancers [7], small, round, blue-cell cancers =-=[8]-=-, and hereditary breast cancer [9]. The next step is to dig deeper and understand the underlying mechanisms and the functions of genes in health and disease. One approach is to model the genetic regul... |

264 | Reveal: a general reverse engineering algorithm for inference of genetic network architectures
- Liang, Fuhrman, et al.
- 1998
(Show Context)
Citation Context ...ancer, it is necessary to be able to identify the networks from real experimental data. Much recent work on Boolean networks has focused on identifying the network structure from gene expression data =-=[58, 59, 60, 61, 62, 63, 64, 65, 66]-=-. At the same time, a large body of related work in computational learning theory [10, 67] has addressed very similar problems, namely, learning or inferring Boolean functions from examples of their i... |

257 | Probabilistic Boolean networks: A rule-based uncertainty model for gene regulatory networks
- Shmulevich, Dougherty, et al.
- 2002
(Show Context)
Citation Context .... Boolean networks are structurally simple yet dynamically complex and have yielded insights into the overall behavior of large genetic networks [22, 31, 32, 33]. Let us give an example borrowed from =-=[34]-=-, showing the logical representation of cell cycle regulation. This process of cellular growth and division is highly regulated. A disbalance in this process results in unregulated cell growth in dise... |

235 |
High density synthetic oligonucleotide arrays
- LIPSHUTZ, FODOR, et al.
- 1999
(Show Context)
Citation Context ...development of high-throughput genomic and proteomic technologies is empowering researchers in the collection of broad-scope gene information. The advent of cDNA microarrays and oligonucleotide chips =-=[1, 2, 3, 4, 5]-=-, which facilitate large-scale surveys of gene expression, has incited much interdisciplinary scientific activity. The diagnostic potential of gene expression data has already been observed. For examp... |

207 | Expression profiling using microarrays fabricated by an ink-jet oligonucleotide synthesizer - Hughes, Mao, et al. - 2001 |

205 | Tissue classification with gene expression profiles
- Ben-Dor
- 2000
(Show Context)
Citation Context ...on data has already been observed. For example, cancer classification using a variety of methods has been used to exploit the class-separating power of expression data: leukemias [6], various cancers =-=[7]-=-, small, round, blue-cell cancers [8], and hereditary breast cancer [9]. The next step is to dig deeper and understand the underlying mechanisms and the functions of genes in health and disease. One a... |

205 |
The logical analysis of continuous, nonlinear biochemical control networks
- Glass, Kauffman
- 1973
(Show Context)
Citation Context ...Why Boolean? The model system that has received, perhaps, the most attention, not only from the biology community, but also in physics, is the Boolean Network model, originally introduced by Kauffman =-=[16, 17, 19, 18]-=-. Good reviews of this model can be found in [20, 21, 22]. In this model, gene expression is quantized to only two levels: ON and OFF. The expression level (state) of each gene is functionally related... |

205 | Identification of genetic networks from a small number of gene expression patterns under the Boolean network model
- Akutsu, Miyano, et al.
- 1999
(Show Context)
Citation Context ...ancer, it is necessary to be able to identify the networks from real experimental data. Much recent work on Boolean networks has focused on identifying the network structure from gene expression data =-=[58, 59, 60, 61, 62, 63, 64, 65, 66]-=-. At the same time, a large body of related work in computational learning theory [10, 67] has addressed very similar problems, namely, learning or inferring Boolean functions from examples of their i... |

205 |
Towards a General Theory of Adaptive Walks on Rugged Landscapes
- Kauffman, Levin
- 1987
(Show Context)
Citation Context ...(a small) probability p, independently of other genes. In the Boolean setting, this is represented by a flip of value from 1 to 0 or vice versa. This idea was already considered by Kauffman and Levin =-=[92, 20] and-=- corresponds to the bit-flipping mutation operator in NK Landscapes. This also corresponds to the mutation operator in genetic algorithms [93]. This type of ‘randomization’, namely allowing genes ... |

192 | Computational limitations on learning from examples
- Pitt, Valiant
- 1988
(Show Context)
Citation Context ...sign of Boolean classifiers forms the core of many data-mining and knowledgediscovery algorithms [67, 68]. For example, the well-known consistency problem represents a search for a rule from examples =-=[69, 70, 71]. That i-=-s, given some sets T and F of “true” and “false” vectors, respectively, the aim is to discover a Boolean function f that takes on the value 1 for all vectors in T and the value 0 5 In a Boolea... |

182 |
At Home in the Universe
- Kauffman
- 1995
(Show Context)
Citation Context ...m order to chaos is a phase transition. In this regime, the number of attractors and the cycle lengths are proportional to powers of n, wheren is the total number of genes. As Stuart Kauffman puts it =-=[54], -=-“a living system must firststrikeaninternalcompromisebetween malleability and stability. To survive in a variable environment, it must be stable to be sure, but not so stable that it remains forever... |

149 | Genomic Cis-Regulatory Logic: Experimental and Computational Analysis of a Sea Urchin
- Yuh, Bolouri, et al.
- 1998
(Show Context)
Citation Context ... is functionally related to the expression states of some other genes, using logical rules. Computational models that reveal these logical interrelations have since then been successfully constructed =-=[23, 24, 25, 26]-=-. Before we go into the formal definitions of the model, it may be useful to 2spause and ask several general, but fundamental questions: 1. To what extent do such models represent reality? 2. Do we ha... |

128 |
Cellular Automata
- Codd
- 1968
(Show Context)
Citation Context ...filters, where the set of root signals represent the associative memory of the filter. Similar work showed that cellular automata, which are special cases of Boolean networks, can process information =-=[46]-=- and are able to perform computations, such as density classification [47, 48]. With the pioneering work of John von Neumann, biology was one of the first disciplines that considered using cellular au... |

127 |
Dynamical behaviour of biological regulatory networks–I. Biological role of feedback loops and practical use of the concept of the loopcharacteristic state
- Thomas
- 1995
(Show Context)
Citation Context ...wered within the seemingly simplistic Boolean formalism. Boolean networks are structurally simple yet dynamically complex and have yielded insights into the overall behavior of large genetic networks =-=[22, 31, 32, 33]-=-. Let us give an example borrowed from [34], showing the logical representation of cell cycle regulation. This process of cellular growth and division is highly regulated. A disbalance in this process... |

113 | Evolving cellular automata to perform computations: Mechanisms and impediments
- Mitchell, Crutchfield, et al.
(Show Context)
Citation Context ... the filter. Similar work showed that cellular automata, which are special cases of Boolean networks, can process information [46] and are able to perform computations, such as density classification =-=[47, 48]-=-. With the pioneering work of John von Neumann, biology was one of the first disciplines that considered using cellular automata for describing and simulating self-reproduction [49]. 3.2 Cell Differen... |

110 |
Computational Learning Theory
- Anthony, Biggs
- 1992
(Show Context)
Citation Context ...ctives. First, we aim to discover and understand the underlying gene regulatory mechanisms by means of inferring them from data. This generally falls within the scope of computational learning theory =-=[10]-=- or system identification [11]. Second, by using the inferred model, we endeavor to make useful predictions by mathematical analysis and computer simulations. There is a natural order to these two obj... |

105 |
The Nature of Statistical Learning Theory, 2nd ed
- Vapnik
- 2000
(Show Context)
Citation Context ...istic) predictor under uncertainty. Although reasoning under uncertainty is not a new problem and has been extensively studied in the artificial intelligence and pattern recognition communities [87], =-=[88]-=-, it nonetheless presents a problem when the uncertainty cannot be reliably estimated. Without doing so, we cannot know how well the designed predictor generalizes over the population. In other words,... |

103 |
Inferring qualitative relations in genetic networks and metabolic pathways
- Akutsu, Miyano, et al.
(Show Context)
Citation Context ...ancer, it is necessary to be able to identify the networks from real experimental data. Much recent work on Boolean networks has focused on identifying the network structure from gene expression data =-=[58, 59, 60, 61, 62, 63, 64, 65, 66]-=-. At the same time, a large body of related work in computational learning theory [10, 67] has addressed very similar problems, namely, learning or inferring Boolean functions from examples of their i... |

100 |
Computational studies of gene regulatory networks: in numero molecular biology
- Hasty, McMillen, et al.
- 2001
(Show Context)
Citation Context ...models, logical networks, Petri nets, graph-based models, grammars, and process algebras. There have been a number of excellent survey papers on modeling and simulation of genetic regulatory networks =-=[12, 13, 14]-=- as well as a recent book [15]. 2 Why Boolean? The model system that has received, perhaps, the most attention, not only from the biology community, but also in physics, is the Boolean Network model, ... |

99 |
It’us a noisy business! Genetic regulation at the nanomolar scale. Trends Genet
- McAdams, Arkin
- 1999
(Show Context)
Citation Context ... Furthermore, genetic regulation exhibits considerable uncertainty on the biological level. Indeed, evidence suggests 3sthat this type of “noise” is in fact advantageous in some regulatory mechani=-=sms [30]-=-. Thus, from a practical standpoint, limited amounts of data and the noisy nature of the measurements can make useful quantitative inferences problematic and a coarse-scale qualitative modeling approa... |

93 | Discovery of regulatory interactions through perturbation: Inference and experimental design. Pacific Symposium on Biocomputing
- Ideker, Thorsson, et al.
- 2000
(Show Context)
Citation Context |

88 |
expression profiling, genetic networks, and cellular states: an integrating concept for tumorigenesis and drug discovery
- Huang, Gene
- 1999
(Show Context)
Citation Context ...gene regulatory networks, can be caused by mutations of the “wiring” of certain genes, thus permanently altering the state-transition structure and, consequently, the long-run behavior of the netw=-=ork [21]-=-. Therefore, it is prudent to develop a methodology for altering the steadystate probabilities of certain states or sets of states with minimal modifications to the rule-based structure. The motivatio... |

85 |
Statistical signal processing
- Scharf
- 1990
(Show Context)
Citation Context ...timal predictor of relative to some error measure . For example, in the case of mean-square error (MSE) estimation, it is well known that the optimal predictor is the conditional expectation of given =-=[84]-=-. Let be the optimal error achieved by . Then, the COD for relative to is defined as where is the error of the best (constant) estimate of in the absence of any conditional variables. It is easily see... |

76 |
Cause-Effect Relationships and Partially Dened Boolean Functions
- Crama, Hammer, et al.
- 1988
(Show Context)
Citation Context ...ient determination of Boolean formulae from examples. This type of induction of Boolean logic or the design of Boolean classifiers forms the core of many data-mining and knowledgediscovery algorithms =-=[67, 68]. For ex-=-ample, the well-known consistency problem represents a search for a rule from examples [69, 70, 71]. That is, given some sets T and F of “true” and “false” vectors, respectively, the aim is to... |

74 |
Morphological Image Processing
- Lotufo, Audigier, et al.
(Show Context)
Citation Context ...median filters, stack filters, and morphological filters [37]–[42]. A key instance of root signals occurs in the case of idempotent filters, which play a central role in mathematical morphology [43]=-=, [44]-=-. Consider a binary-valued one-dimensional (1-D) signal of arbitrary length. Suppose a window of length is sliding across this signal. At every location of the window, the contents inside the window a... |

69 |
Stack Filters
- Wendt, Coyle, et al.
- 1986
(Show Context)
Citation Context ...ch like the frequencies that are passed by a linear filter. Root signals have been studied extensively for different types of filters, such as median filters, stack filters, and morphological filters =-=[37, 38, 39, 40, 41, 42]-=-. Consider a binary-valued 1-D signal of arbitrary length. Suppose a window of length n =2m +1is sliding across this signal. At every location of the window, the contents inside the window are used as... |

66 | Identification of gene regulatory networks by strategic gene disruptions and gene overexpressions
- Akutsu, Kuhara, et al.
- 1998
(Show Context)
Citation Context |

66 |
System Identification: Theory for the User. Upper Saddle
- Ljung
- 1999
(Show Context)
Citation Context ...ver and understand the underlying gene regulatory mechanisms by means of inferring them from data. This generally falls within the scope of computational learning theory [10] or system identification =-=[11]-=-. Second, by using the inferred model, we endeavor to make useful predictions by mathematical analysis and computer simulations. There 1778 PROCEEDINGS OF THE IEEE, VOL. 90, NO. 11, NOVEMBER 2002sis a... |

64 |
Computational Modeling of Genetic and Biochemical Networks, ch. A logical model of cis-regulatory control in eukaryotic system
- Yuh, Bolouri, et al.
- 2001
(Show Context)
Citation Context ...graph-based models, grammars, and process algebras. There have been a number of excellent survey papers on modeling and simulation of genetic regulatory networks [12, 13, 14] as well as a recent book =-=[15]-=-. 2 Why Boolean? The model system that has received, perhaps, the most attention, not only from the biology community, but also in physics, is the Boolean Network model, originally introduced by Kauff... |

64 |
The Nature of Statistical Learning Theory, Second Edition
- Vapnik
- 1999
(Show Context)
Citation Context ...terministic) predictor under uncertainty. Although reasoning under uncertainty is not a new problem and has been extensively studied in the artificial intelligence and pattern recognition communities =-=[85, 86]-=-, it nonetheless presents a problem when the uncertainty cannot be reliably estimated. Without doing so, we cannot know how well the designed predictor generalizes over the population. In other words,... |

64 |
A methodology for solving Markov models of parallel systems
- Plateau, Fourneau
- 1991
(Show Context)
Citation Context ... them is “ready” to make the next transition should it be selected. Very similar ideas have been used in the analysis of parallel and distributed systems, using so-called stochastic automata netwo=-=rks [89]-=-. It is easy to see that the state space of a PBN is the same as of a standard Boolean network. Namely, there are 2 n possible states, each represented by a binary vector of length n. The difference i... |

64 | Gene Perturbation and Intervention in Probabilistic Boolean Networks
- Shmulevich, Dougherty, et al.
- 2002
(Show Context)
Citation Context ... is given by with probability with probability where is component-wise addition modulo 2 and , is the transition function representing a possible realization of the entire PBN, as discussed above. In =-=[96]-=-, an explicit formulation of the state-transition probabilities in terms of the Boolean functions and the probability of perturbation , was derived. It is fairly easy to show [96] that, for , the Mark... |

62 |
A theoretical analysis of the properties of median filters
- Gallagher, Wise
- 1981
(Show Context)
Citation Context ...ch like the frequencies that are passed by a linear filter. Root signals have been studied extensively for different types of filters, such as median filters, stack filters, and morphological filters =-=[37, 38, 39, 40, 41, 42]-=-. Consider a binary-valued 1-D signal of arbitrary length. Suppose a window of length n =2m +1is sliding across this signal. At every location of the window, the contents inside the window are used as... |

62 | Development of a system for the inference of large scale genetic networks - Maki, Tominaga, et al. - 2001 |

58 | Genomic regulation modeled as a network with basins of attraction
- Wuensche
- 1998
(Show Context)
Citation Context ...nctions assigned to them and this process is then repeated. The artificial synchrony simplifies computation while preserving the qualitative, generic properties of global network dynamics [21], [20], =-=[32]-=-. It is clear that the dynamics of the network are completely determined by (1). Let us give an example. Consider a Boolean network consisting of five genes with the corresponding Boolean functions gi... |

56 | Binary analysis and optimization-based normalization of gene expression data
- Shmulevich, Zhang
- 2002
(Show Context)
Citation Context ...ith a Boolean formalism, it is prudent to test whether or not meaningful biological information can be extracted from gene expression data entirely in the binary domain. This question was taken up in =-=[27]-=-. We reasoned that if the genes, when quantized to only two levels (1 or 0), would not be informative in separating known sub-classes of tumors, then there would be little hope for Boolean modeling of... |

55 |
Hidden Order: How Adaptation Builds Complexity (Helix Books
- Holland
- 1995
(Show Context)
Citation Context ...tion and Cellular Functional States Boolean networks qualitatively reflect the nature of complex adaptive systems in that they are “systems composed of interacting agents described in terms of rules=-=” [51]-=-. A central concept in dynamical systems is that of structural stability, which is the persistent behavior of a system under perturbation. Structural stability formally captures the idea of behavior t... |