Systems and methods for reverse engineering models of biological networks

• US 20060293873A1
• Filed: 03/05/2003
• Published: 12/28/2006
• Est. Priority Date: 03/06/2002
• Status: Abandoned Application

• Current Assignee: Trustees of Boston University
• Original Assignee: Trustees of Boston University
• Inventors: Yeung, Man Kit Stephen, di Bernardo, Diego, Collins, James J., Tegner, Jesper, Gardner, Timothy S.
• Application Number:
  US10/506,734
• Publication Number:
  US 20060293873A1
• Time in Patent Office: Days
• US Class Current: 703/11
• CPC Class Codes: G16B 20/00 ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
• CPC Class Codes: G16B 40/00 ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
• CPC Class Codes: G16B 5/00 ICT specially adapted for modelling or simulations in systems biology, e.g. gene-regulatory networks, protein interaction networks or metabolic networks
• CPC Class Codes: G16B 5/20 Probabilistic models

Abstract

The present invention provides methods and accompanying computer-based systems and computer-executable code stored on a computer-readable medium for constructing a model of a biological network. The invention further provides methods for performing sensitivity analysis on a biological network and for identifying major regulators of species in the network and of the network as a whole. In addition, the invention provides methods for identifying targets of a perturbation such as that resulting from exposure to a compound or an environmental change. The invention further provides methods for identifying phenotypic mediators that contribute to differences in phenotypes of biological systems.

Claims

1. A method of constructing a model of a biological network comprising steps of:

providing a biological system or a plurality of biological systems, each biological system comprising a biological network comprising a plurality of biochemical species having activities;

perturbing the activity of at least one of the biochemical species, thereby causing a response in the biological network;

allowing the biological network to reach a steady state;

determining the response of at least one of the biochemical species in the biological network; and

estimating parameters of the model.

2. The method of claim 1, wherein the model comprises a set of differential equations or a set of difference equations that represent evolution over time of the activities of at least one of the biochemical species.

3. The method of claim 1, wherein the model comprises a polynomial approximation of a set of differential equations or a set of difference equations that represent evolution over time of the activities of at least one of the biochemical species.

4. The method of claim 1, wherein the model comprises a linear approximation of a set of differential equations or a set of difference equations that represent evolution over time of the activities of at least one of the biochemical species.

5. The method of claim 2, 3, or 4, wherein the differential or difference equations are nonlinear.

6. The method of claim 1, wherein determining the response of a biochemical species comprises measuring the activity of the biochemical species.

7. The method of claim 1, wherein the response comprises an alteration in the activity of one or more of the biochemical species in the biological network or no alteration in any of the biochemical species.

8. The method of claim 1, wherein the perturbing step comprises:

applying a perturbation to a different biochemical species in the biological network in each of at least one of the biological systems, each biological system comprising a cell or a population of cells, and wherein the determining step comprises determining the response of at least one of the biochemical species in the biological network in each of at least one of the biological systems after allowing the biological network to reach a steady state.

9. The method of claim 1, wherein the perturbing step comprises:

applying a perturbation to one or more biochemical species in the biological network in each of at least one of the biological systems, each biological system comprising a cell or a population of cells, and wherein the determining step comprises determining the response of at least one of the biochemical species in the biological network in each of at least one of the biological systems after allowing the biological network to reach a steady state.

10. The method of claim 8 or 9, wherein a single biochemical species in the biological network in each biological system is perturbed.

11. The method of claim 8 or 9, wherein multiple biochemical species in each biological system are perturbed simultaneously.

12. The method of claim 8 or 9, wherein each of the biochemical species in the biological network is perturbed in at least one of the biological systems.

13. The method of claim 8 or 9, wherein each of the biological species in the biological network is perturbed in exactly one of the biological systems.

14. The method of claim 8 or 9, wherein less than 100% of the biochemical species in the biological network are perturbed.

15. The method of claim 1, wherein the perturbing step comprises:

applying a perturbation to one or more biochemical species in the biological network in a biological system comprising a cell or a population of cells, and wherein the determining step comprises determining the response of at least one of the biochemical species in the biological network after allowing the biological network to reach a steady state; and

repeating the applying and determining steps for each of at least one of the biochemical species in the biological network.

16. The method of claim 15, wherein a single biochemical species in the biological network is perturbed in each applying step.

17. The method of claim 15, wherein multiple biochemical species are perturbed simultaneously in each applying step.

18. The method of claim 15, wherein each of the biochemical species in the biological network is perturbed in at least one of the applying steps.

19. The method of claim 15, wherein each of the biological species in the biological network is perturbed in exactly one of the applying steps.

20. The method of claim 15, wherein less than 100% of the biochemical species in the biological network are perturbed.

21. The method of claim 1, wherein the biological system is a cell or a population of cells.

22. The method of claim 21, wherein the cell or population of cells is prokaryotic or eukaryotic.

23. The method of claim 1, wherein the activity of a biochemical species is its expression level.

24. The method of claim 1, wherein the biochemical species comprise at least one of genes, deoxyribonucleic acids, ribonucleic acids, polypeptides, and metabolites, and wherein the activity of a biochemical species is its amount, concentration, maximum catalytic rate or expression level.

25. The method of claim 1, wherein the step of estimating parameters comprises steps of:

selecting a fitness function; and

either computing the values of the parameters that optimize the fitness function;

or

(i) selecting a search procedure; and

(ii) applying the selected search procedure so as to identify the values of the parameters that optimize the selected fitness function.

26. The method of claim 25, wherein determining the response of a biochemical species comprises measuring the activity of the biochemical species, and wherein the fitness function compares measured values of the perturbations applied in the perturbing step with predictions of the measured values of the perturbations.

27. The method of claim 26, wherein the predictions are obtained by using the measured activity values, selected values of the parameters, and the model to calculate values of the perturbations that would produce the measured activities, given the selected values of the parameters and the model.

28. The method of claim 25, wherein the fitness function is selected from the group consisting of:

a minimum total square error function, a maximum square error function, and a total absolute error function.

29. The method of claim 25, wherein the search procedure is selected from the group consisting of:

a Simplex algorithm, a gradient descent algorithm, and a simulated annealing algorithm.

30. The method of claim 25, further comprising the step of:

imposing a constraint on the biological network so as to limit values that may be used to optimize the fitness function.

31. The method of claim 30, wherein the search procedure is selected from the group consisting of:

a Simplex algorithm, a gradient descent algorithm, and a simulated annealing algorithm.

32. The method of claim 30, wherein the constraint comprises one or more constraints selected from the group consisting of:

(i) fixing the number of regulatory inputs to each biochemical species;

(ii) minimizing the number of non-zero parameters;

(iii) restricting parameters to discrete values; and

(iv) requiring parameters that provide a dynamically stable model.

33. The method of claim 32, further comprising the step of:

estimating the number of regulatory inputs to each biochemical species, wherein the constraint comprises fixing the number of regulatory inputs to each biochemical species.

34. The method of claim 32, wherein the constraint comprises selecting the maximum number of regulatory inputs to each biochemical species and, wherein the search procedure comprises:

(a) generating all putative network structures including one or more regulatory inputs per biochemical species, but not more regulatory inputs than the maximum number of regulatory inputs;

(b) calculating or searching for parameters that optimize a chosen fitness function for each putative network structure; and

(c) selecting as a solution whichever of the putative networks of step (b), comprising a network structure and parameters, optimizes the fitness function.

35. The method of claim 34, wherein the fitness function compares measured values for perturbations applied in the perturbing step with predictions of the measured values of the perturbations.

36. The method of claim 35, wherein the predictions are obtained by using the measured activity values, selected values of the parameters, and the model to calculate values of the perturbations that would produce the measured activities, given the selected values of the parameters and the model.

37. The method of claim 32, wherein the constraint comprises fixing the number of regulatory inputs to each biochemical species;

and wherein the search procedure comprises steps of;

(a) generating one or more putative network structures including one or more input connections per biochemical species;

(b) calculating or searching for parameters that optimize a chosen fitness function for each putative network structure;

(c) selecting one or more of the networks of step (b), each putative network comprising a structure and parameters, with optimal fitness as determined by the fitness function;

(d) determining whether any of the putative networks selected in part (c) satisfies some chosen stop criterion;

(e) if the stop criterion is met, stopping the search and selecting one or more putative network structures and parameters; and

(f) if the stop criterion is not met, generating one or more variants of the network structures selected in step (c) and returning to step (b).

38. The method of claim 37, wherein the fitness function compares measured values for perturbations applied in the perturbing step with predictions of the measured values of the perturbations.

39. The method of claim 38, wherein the predictions are obtained by using the measured activity values, selected values of the parameters, and the model to calculate values of the perturbations that would produce the measured activities, given the selected values of the parameters and the model.

40. The method of claim 37, wherein the searching in step (b) is performed using a search procedure selected from the group consisting of:

a Simplex algorithm, a gradient descent algorithm, and a simulated annealing algorithm.

41. The method of claim 37, wherein the search procedure is selected from the group consisting of:

Forw-reest-K, Forw-TopD-reest-K, Forw-Float-K, Back-K, Back-reest-K, Genalg-SteadyState-K, Genalg-Gen-K, and Exhaustive-K.

42. The method of claim 37, wherein the stop criterion requires that the putative network attains a predetermined level of fitness, that the putative network comprises a selected number of regulatory inputs, or that the change in the level of fitness between subsequent iterations of the steps (b) and (c) is less than a predetermined amount.

43. The method of claim 32, wherein the constraint comprises minimizing the number of nonzero parameters.

44. The method of claim 43, wherein the fitness function is the total square error function and the search procedure identifies auxiliary parameters that minimize a function consisting of the sum of (i) the minimum length vector of the parameters that minimize the total square error fitness function and (ii) the product of (a) the auxiliary parameters and (b) a matrix of vectors spanning the nullspace of a matrix comprising the measured activities.

45. The method of claim 1, wherein the estimated parameters are considered random variables;

and comprising the additional step of estimating the probability density function for each estimated parameter.

46. The method of claim 45, wherein estimating the probability density function for each estimated parameter comprises estimating one or more of the first, second, third or higher moments of the probability density function.

47. The method of claim 46, wherein one or more of the first, second, third or higher moments of the probability density function for each parameter are estimated using the measured activity values and the measured perturbation values.

48. The method of claim 47, wherein the estimated first and second moments of the probability density functions of the estimated parameters are used to calculate the statistical significance of the one or more of the estimated parameters.

49. The method of claim 48, wherein the statistical significance of one or more of the estimated parameters are calculated using one or more tests selected from the group consisting of the z-test, the t-test, and the chi-squared-test.

50. A method of performing sensitivity analysis on a biological network comprising steps of:

generating or providing a model of the biological network according to the method of claim 1;

and

determining the sensitivity of the activities of a first set of one or more species in the network to a change in the activities of a second set of one or more species in the network using the model.

51. The method of claim 50, further comprising the step of:

identifying the second set of species as a major regulator of the first set of species if the sensitivity of the first set of species to a change in the activities of the second set of species meets a predefined criterion.

52. The method of claim 50, wherein the determining step comprises:

inverting a matrix of parameters of the model, wherein the parameters quantify regulatory relationships between species in the model, thereby obtaining a gain matrix, wherein each column or row of the gain matrix represents the sensitivity of species in the model to a perturbation of one of the species.

53. The method of claim 50, wherein the sensitivity of the activities a first set of biochemical species to a change in the activities of a second set of biochemical species is a measure of the change in activities of the first set of species in response to a change in activities of the second set of species.

54. The method of claim 53, wherein the measure is a quantitative measure.

55. The method of claim 54, wherein the quantitative measure is the mean percentage change in activities of the first set of species in response to a unit change in activities of the second set of species.

56. The method of claim 51, wherein the predefined criterion is a requirement that sensitivity of the activities of at least one species in the first set of species to a change in the activities of the second set of activities is statistically different from zero.

57. The method of claim 51, wherein the predefined criterion is a requirement that the sensitivity of the activities of at least one species in the first set of species to a change in the activities of the second set of activities exceeds a predetermined value.

58. The method of claim 51, wherein the predefined criterion is a requirement that the sensitivity of the activities of the first set of species to a change in the activities of the second set of species is greater than the sensitivity of the first set of species to a change in the activities of a third set of one or more species.

59. A method of identifying a target of a perturbation comprising steps of:

providing a biological system comprising a biological network comprising a plurality of biochemical species having activities;

providing or generating a model of the biological system constructed according to the method of claim 1;

perturbing one or more biochemical species in the network;

allowing the biological network to reach a steady state;

determining the response of at least one of the biochemical species in the biological network to the compound; and

calculating predicted perturbations of biochemical species in the biological network that would be expected to yield the determined responses according to the model.

60. The method of claim 59, wherein the perturbing step comprises contacting the biological system with a compound, thereby causing a response in the biological network, and wherein the target is a target of the compound.

61. The method of claim 59, further comprising the step of:

identifying a biochemical species as a target of the perturbation if the predicted perturbation to that biochemical species meets a predefined criterion.

62. The method of claim 61, wherein the predefined criterion is a requirement that the strength of the predicted perturbation to the biochemical species exceeds a predetermined value.

63. The method of claim 61, wherein the predefined criterion is a requirement that the strength of the predicted perturbation is identified as statistically significant.

64. The method of claim 63, wherein apredicted perturbation is identified as statistically significant by using a statistical test selected from the group consisting of the z-test, the t-test, and the chi-squared-test.

65. The method of claim 64, wherein the statistical test is used with estimates of the first and second moments of the probability density functions of the predicted perturbations, wherein the estimates of the first and second moments are calculated from measured values of the responses of the biochemical species and measured values of the perturbations applied in the perturbing step.

66. A method for identifying phenotypic mediators comprising steps of:

comparing parameters of models of biological networks for a plurality of biological systems, wherein the models are generated according to the method of claim 1, and wherein the biological networks comprise overlapping or substantially identical sets of biochemical species; and

identifying biochemical species for which associated parameters differ between the models as candidate phenotypic mediators.

67. The method of claim 66, wherein one or more of the biological systems display differences in one or more properties.

68. The method of claim 66, wherein the properties include the steady-state activities of the biochemical species of the biological system, the phenotype of the biological system, and the genotype of the biological system.

69. The method of claim 66, wherein a species is identified as a phenotypic mediator if the difference between the parameters for that species in some or all of the models satisfies a predefined criterion.

70. The method of claim 69, wherein the predefined criterion is a requirement that the difference exceeds a predefined value.

71. The method of claim 69, wherein the predefined criterion is a requirement that the difference achieves a particular level of statistical significance.

72. A computer system for constructing a model of a biological network, the computer system comprising:

memory that stores a program comprising computer-executable process steps; and

a processor which executes the process steps so as to construct a model of a biological network, the model comprising an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network.

73. The computer system of claim 72, wherein the process steps estimate parameters of and select a structure for a model of a biological network.

74. The computer system of claim 73, wherein the process steps comprise steps of:

(a) selecting or receiving a selection of a fitness function; and

(b) either computing the values of the parameters that optimize the fitness function;

or

(i) selecting or receiving a selection of a search procedure; and

(ii) applying the search procedure so as to identify the values of the parameters that optimize the fitness function.

75. The computer system of claim 74, wherein the search procedure comprises:

(a) generating all putative network structures including one or more regulatory inputs per biochemical species, but not more regulatory inputs than the maximum number of regulatory inputs;

(b) calculating or searching for parameters that optimize a chosen fitness function for each putative network structure; and

(c) selecting as a solution whichever of the putative networks of step (b), comprising a network structure and parameters, optimizes the fitness function.

76. A computer system for performing sensitivity analysis of a biological network, the computer system comprising:

memory that stores a program comprising computer-executable process steps; and

a processor which executes the process steps so as to

(i) generate or receive a model of a biological network, the model comprising an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network; and

(ii) determine the sensitivity of the activities of a first set of one or more species in the network to a change in the activities of a second set of one or more species in the network using the model.

77. A computer system for identifying a target of a perturbation comprising

memory that stores a program comprising computer-executable process steps;

and

a processor which executes the process steps so as to

(i) construct or receive a model of a biological network, the model comprising an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network; and

(ii) receive data comprising responses of at least one of the biochemical species to the perturbation; and

(iii) calculate predicted perturbations of biochemical species in the biological network that would be expected to yield the determined responses according to the model.

78. A computer system for identifying a phenotypic mediator, the computer system comprising:

memory that stores a program comprising computer-executable process steps; and

a processor which executes the process steps so as to

(i) compare parameters of models of biological networks for a plurality of biological systems, wherein the models comprise an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network, and wherein the biological networks comprise overlapping or substantially identical sets of biochemical species; and

(ii) identify one or more biochemical species for which associated parameters differ between the models as candidate phenotypic mediators.

79. Computer-executable process steps stored on a computer-readable medium, the computer-executable process steps to construct a model of a biological network, the computer-executable process steps comprising:

code to construct a model of a biological network, the model comprising an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network.

80. Computer-executable process steps as set forth in claim 79, wherein the code comprises code to estimate parameters of and select a structure for a model of a biological network.

81. Computer-executable process steps as set forth in claim 80, wherein the code to estimate parameters and select a structure comprises

code to select or receive a selection of a fitness function;

and

code to perform either or both of (a) and (b), where (a) and (b) are;

(a) code to compute the values of the parameters that optimize the fitness function;

(b) code to

(i) select or receive a selection of a search procedure; and

(ii) apply the selected search procedure so as to identify the values of the parameters that optimize the selected fitness function.

82. Computer-executable process steps as set forth in claim 80, wherein the search procedure comprises steps of:

(a) generating all putative network structures including one or more regulatory inputs per biochemical species, but not more regulatory inputs than the maximum number of regulatory inputs;

(b) calculating or searching for parameters that optimize a chosen fitness function for each putative network structure; and

(c) selecting as a solution whichever of the putative networks of step (b), comprising a network structure and parameters, optimizes the fitness function.

83. Computer-executable process steps stored on a computer-readable medium, the computer-executable process steps to perform sensitivity analysis of a biological network, the computer-executable process steps comprising:

(i) code to generate or receive a model of a biological network, the model comprising an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network; and

(ii) code to determine the sensitivity of the activities of a first set of one or more species in the network to a change in the activities of a second set of one or more species in the network using the model.

84. Computer-executable process steps stored on a computer-readable medium, the computer-executable process steps to identify a target of a perturbation, the computer-executable process steps comprising:

(i) code to estimate parameters of and select a structure for a model of a biological network or to receive estimated parameters of and a structure for a model of a biological network, the model comprising an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network;

(ii) code to receive data comprising responses of at least one of the biochemical species to the perturbation; and

(iii) code to calculate predicted perturbations of biochemical species in the biological network that would be expected to yield the determined responses according to the model.

85. Computer-executable process steps stored on a computer-readable medium, the computer-executable process steps to identify a phenotypic mediator, the computer-executable process steps comprising:

(i) code to compare parameters of models of biological networks for a plurality of biological systems, wherein the models comprise an approximation to a set of differential equations or a set of difference equations that represent evolution over time of activities of at least one biochemical species in a biological network, and wherein the biological networks comprise overlapping or substantially identical sets of biochemical species; and

(ii) code to identify one or more biochemical species for which associated parameters differ between the models as candidate phenotypic mediators.