Statistical deconvoluting of mixtures

• US 6,434,542 B1
• Filed: 01/19/2000
• Issued: 08/13/2002
• Est. Priority Date: 04/17/1997
• Status: Expired due to Fees

• Current Assignee: Smithkline Beecham Corporation (GSK plc)
• Original Assignee: Smithkline Beecham Corporation (GSK plc)
• Inventors: Lambert, Christophe Gerard, Young, Sidney Stanley, Farmen, Mark W., Rusinko, Andrew R. III
• Primary Examiner(s): Black, Thomas
• Assistant Examiner(s): Holmes, Michael B.
• Application Number:
  US09/403,163
• Time in Patent Office: 937 Days
• Field of Search: 706/45, 711/202, 703/4, 707/532, 345/762, 514/12, 435/69.4
• US Class Current: 706/45
• CPC Class Codes: B01J 2219/0054 Means for coding or tagging the apparatus or the reagents
• CPC Class Codes: B01J 2219/00702 Processes involving means for analysing and characterising the products
• CPC Class Codes: C40B 70/00 Tags or labels specially adapted for combinatorial chemistry or libraries, e.g. fluorescent tags or bar codes
• CPC Class Codes: G06F 2216/03 Data mining
• CPC Class Codes: G16B 20/00 ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
• CPC Class Codes: G16B 20/20 Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
• 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: G16C 20/30 Prediction of properties of chemical compounds, compositions or mixtures
• CPC Class Codes: G16C 20/70 Machine learning, data mining or chemometrics

Abstract

Statistical classification of activities of molecules is a computer implemented methodology of QSAR employing visualization of molecular features and statistical techniques for correlating features of molecules with their observed biological properties. Each molecule is described by noting the presence (1) or absence (0) of a feature of interest. The identification of specific features coded by 1'"'"'s or 0'"'"'s is accomplished by recursive partitioning. The data sets are planned or unplanned. The method is also applicable to classification of individuals in biological populations on the basis of their genetic makeup.

Claims

1. A computer-based method of encoding features of data objects, and of identifying and correlating individual said features to a response characteristic that is a trait of interest of the data object, applicable to data objects in a data set that is characterized in being a mixture of data object classes, each data object class containing one or more of said data objects, and wherein multiple data objects present a same or similar value of the trait of interest, but classes of data objects produce the response characteristic that is a trait of interest through different underlying mechanisms,

comprising the steps of:

(a) assembling a set of descriptors and converting said set of descriptors into the form of a bit string such that each descriptor reflects the presence or absence of any given potentially useful feature of interest in a data object of interest;

(b) examining each data object for presence or absence of each of said descriptors;

(c) assembling the results of step (b) into a vector for each data object, noting the presence or absence of each feature of interest in said data object;

(d) assembling all vectors generated in step (c) into a matrix with each row of the matrix corresponding to a data object and each column corresponding to a feature of interest;

(e) dividing the data in said matrix into two daughter sets on the basis of presence or absence of a given feature of interest from said set of descriptors; and

(f) repeating step (e) until each member of said matrix has been identified in terms of presence or absence of any given feature of interest from said set of descriptors and each of said members has been assigned to a terminal node.

11. The method as claimed in claim 1, including the additional step of assembling a chemical structure data file.

12. The method as claimed in claim 1, including the additional step of assembling biological data pertaining to each chemical mixture or mixture of chemicals and assigning each chemical mixture its biological data.

13. The method as claimed in claim 1, in which said correlation is between presence or absence of one or more chemical descriptors and biological activity of a chemical mixture.

14. The method as claimed in claim 1, in which said correlation is between presence or absence of one or more chemical descriptors and pharmacological activity of a chemical compound.

15. The method as claimed in claim 1, including the additional step of determining structure-activity relationships, such relationships comprising sets of rules defining the sets of features specific to each activity class.

16. The method as claimed in claim 1, in which said descriptor is an atom pair.

17. The method as claimed in claim 1, in which said descriptor is an atom triple.

18. The method as claimed in claim 17, in which said atom triple is a set of three defined atoms in a molecule of interest, each atom defined by element, by spatial relation to each of the other two atoms, and by the type of chemical bond or number of chemical bonds separating them in the molecule.

19. The method as claimed in claim 1, in which said descriptor is a molecular fragment.

20. The method as claimed in claim 1, in which said descriptor is a molecular topological torsion.

21. The method as claimed in claim 1, in which said descriptor is a measure of thermodynamic stability.

22. The method as claimed in claim 1, in which said descriptor is a binary of continuous variable.

23. The method as claimed in claim 1, in which said descriptor is a combination in any order of an atom pair, an atom triple, a molecular fragment, a molecular topological torsion, thermodynamic stability or a binary of a continuous variable.

24. The method as claimed in claim 1, in which each descriptor is an element of a vector in said matrix.

25. The method as claimed in claim 1, in which presence or absence of each feature of interest is represented as a 1 or a 0, respectively.

26. The method as claimed in claim 24, in which said vector is computationally represented as a bit string data file.

27. The method as claimed in claim 26, in which said bit string data file is utilized to computationally create a bit string data file.

28. The method as claimed in claim 26, in which said bit string is computationally compressed into a sparse matrix.

29. The method as claimed in claim 28, in which said sparse matrix is statistically analyzed by recursive partitioning.

30. The method as claimed in claim 29, in which said recursive partitioning is performed by the CART method.

31. The method as claimed in claim 29, in which said recursive partitioning is performed by the FIRM method.

32. The method as claimed in claim 29, in which said recursive partitioning is performed by the C4.5 method.

33. The method as claimed in claim 31, in which said FIRM method is converted from multiway splits to binary splits.

34. The method as claimed in claim 1, including the additional step of selecting the descriptor that optimally divides said rows of said data matrix into two subsets of rows, being either compounds or mixtures of compounds where said feature of interest is present or absent, respectively, and repeating this process through subsequent iterations until all descriptors in said descriptor set have been examined repeatedly and all said rows assigned to terminal nodes.

35. The method as claimed in claim 1, in which the result of said recursive partitioning is graphically represented as a recursive partitioning analysis tree.

36. The method as claimed in claim 1, in which said data objects are discrete compounds.

2. A computer-based apparatus system for allowing a user thereof to encode features of data objects, and to identify and correlate individual said features to a response characteristic that is a trait of interest of the data object, applicable to data objects in a data set that is characterized in being a mixture of data object classes, each data object class containing one or more of said data objects, and wherein multiple data objects present a same or similar trait of interest, but classes of data objects produce the response characteristic that is a trait of interest through different underlying mechanisms, comprising:

(a) input means responsive to operator commands enabling an operator to specify a set of descriptors that are subsequently converted into a bit-string, such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a data object of interest;

(b) storage means for storing the assembled set of (a);

(c) memory means for executing programmed steps that examine each data object for presence or absence of each of said descriptors;

(d) means for assembling the results of (c) into a virtual matrix with each row of the matrix corresponding to an object and each column corresponding to a feature of interest;

(e) means for assigning each data object in said matrix recursively into one of two defined categories on the basis of presence or absence of a given feature of interest from said set of descriptors and repeating such analysis until each member of said mixture has been identified in terms of presence or absence of features of interest from said set of descriptors and assigned to a terminal node; and

(f) output means for visually displaying, using computer graphics, a relationship of said descriptors with said data objects and classes.

3. A computer software system having a set of instructions for controlling a general purpose digital computer in performing a desired function comprising:

a set of instructions formed into each of a plurality of modules, each module comprising;

(a) an input process responsive to operator commands enabling an operator to specify a set of descriptors and convert said descriptors into a bit string such that each descriptor reflects the presence or absence of a potentially useful feature of interest of a data object of interest, wherein each data object is a member of a data set that is characterized in being a mixture of data object classes, each data object class containing one or more of said data objects, and wherein multiple data objects present a same or similar trait of interest, but classes of data objects produce the response characteristic that is a trait of interest through different underlying mechanisms;

(b) a data storage process for storing the assembled set of (a);

(c) a computational process for executing programmed steps that examine each member of said mixture for presence or absence of each of said descriptors;

(d) a computational process for assembling the results of (c) into a vector for each data object and a matrix for all vectors;

(e) a computational process for assigning each data object in said matrix into one of two defined categories on the basis of presence or absence of a given feature of interest from said set of descriptors and repeating such analysis until each member of said mixture has been identified in terms of presence or absence of each feature of interest from said set of descriptors and assigned to a terminal node;

(f) a data storage process; and

(g) an output process for visually displaying, using computer graphics, a relationship of said descriptors with said data objects and classes.

4. A computer-based method of encoding mixture features of planned mixtures or of inadvertent mixtures, or of a combination of planned or inadvertent mixtures, and of identifying and correlating individual said features to a response characteristic of the mixture object, wherein said mixture object is in a data set wherein multiple mixture objects comprising the data set present the same trait of interest through a common underlying mechanism;

comprising the steps of;

(a) assembling a set of descriptors and converting said set of descriptors into the form of a bit string such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a mixture object;

(b) examining each mixture object for presence or absence of each of said descriptors;

(c) assembling the results of step (b) into a vector for each mixture object, noting the presence or absence of each feature of interest in said mixture object;

(d) assembling all vectors generated in step (c) into a matrix with each row corresponding to a mixture object and each column corresponding to a feature of interest;

(e) dividing the mixture objects in said matrix into two defined daughter nodes on the basis of presence or absence of a given feature of interest from said set of descriptors; and

(f) repeating step (e) until each mixture object of said matrix has been identified in terms of presence or absence of given features of interest from said set of descriptors and assigned to a terminal node.

37. The method as claimed in claim 4, in which said data objects are mixtures of discrete compounds.

5. A computer-based apparatus system for allowing a user thereof to encode features of planned mixtures or of inadvertent mixtures, or of a combination of planned or inadvertent mixtures, and to identify and correlate individual said features to a response characteristic of the mixture object, wherein said mixture object is in a data set wherein multiple mixture objects comprising the data set present the same trait of interest through a common underlying mechanism, comprising:

(a) input means responsive to operator commands enabling an operator to specify a set of descriptors that are subsequently converted into a bit string, such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a mixture object of interest;

(b) storage means for storing the assembled set of (a);

(c) memory means for executing programmed steps that examine each mixture object for presence or absence of each of said descriptors;

(d) means for assembling the results of (c) into a virtual matrix with each row corresponding to a mixture object and each column corresponding to a feature;

(e) means for assigning each mixture object in said matrix recursively into one of two defined categories on the basis of presence or absence of a given feature of interest from said set of descriptors and repeating such analysis until each mixture object of said matrix population has been classified in terms of presence or absence of given features of interest from said set of descriptors and assigned to a terminal node; and

(f) output means for visually displaying, using computer graphics, the relationships of said descriptors with said mixture classes and mixture objects.

6. A computer software system having a set of instructions for controlling a general purpose digital computer in performing a desired function comprising:

a set of instructions formed into each of a plurality of modules, each module comprising;

(a) an input process responsive to operator commands enabling an operator to specify a set of descriptors and convert said descriptors into a bit string such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a mixture object of interest, wherein each mixture object is a member of a data set where each mixture object presents a same trait of interest through a common underlying mechanism;

(b) a data storage process for storing the assembled set of (a);

(c) a computational process for executing programmed steps that examine each member object of said data set for presence or absence of each of said descriptors;

(d) a computational process for assembling the results of (c) into a vector for each mixture object and a virtual matrix with each row corresponding to a mixture object and each column corresponding to a feature;

(e) a computational process for analyzing the data in said matrix into one of two defined categories on the basis of presence or absence of a given feature of interest from said set of descriptors and repeating such analysis until each member of said mixture has been identified in terms of presence or absence of each feature of interest from said set of descriptors and assigned to a terminal node;

(f) a data storage process; and

(g) an output process for visually displaying, using computer graphics, a relationship of said descriptors with said mixture objects and classes.

7. A computer-based method of encoding mixture features of planned mixtures or of inadvertent mixtures, or of a combination of planned or inadvertent mixtures, and of identifying and correlating individual said features to a response characteristic that is a trait of interest of the mixture object, wherein said mixture object is in a data set that is characterized in being a mixture of mixture object classes, each class containing one or more of said mixture objects, and wherein multiple mixture objects present a same trait of interest, but classes of mixture objects produce the response characteristic which is a trait of interest through different underlying mechanisms,

comprising the steps of:

(a) assembling a set of descriptors and converting said set of descriptors into the form of a bit string such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a mixture object of interest;

(b) examining each mixture object for presence or absence of each of said descriptors;

(c) assembling the results of step (b) into a vector for each mixture object, noting the presence or absence of each feature in said data object;

(d) assembling all vectors generated in step (c) into a matrix with each row corresponding to a mixture object and each column corresponding to a feature;

(e) dividing the mixture objects in said matrix into two defined daughter nodes on the basis of presence or absence of a given feature of interest from said set of descriptors; and

(f) repeating step (e) until each mixture object of said matrix has been identified in terms of presence or absence of given features of interest from said set of descriptors and assigned to a terminal node.

8. A computer-based apparatus system for allowing a user thereof to encode features of planned mixtures or of inadvertent mixtures, or of a combination of planned or inadvertent mixtures, and to identify and correlate individual said features to a response characteristic that is a trait of interest of the mixture object, applicable to mixture objects in a data set that is characterized in being a mixture of mixture object classes, each class containing one or more of said mixture objects, and wherein multiple mixture objects present a same trait of interest, but classes of mixture objects produce the response characteristic that is a trait of interest through different underlying mechanisms, comprising:

(a) input means responsive to operator commands enabling an operator to specify a set of descriptors that are subsequently converted into a bit string, such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a mixture object of interest;

(b) storage means for storing the assembled set of (a);

(c) memory means for executing programmed steps that examine each mixture object for presence or absence of each of said descriptors;

(d) means for assembling the results of (c) into a virtual matrix with each row corresponding to a mixture object and each column corresponding to a feature;

(e) means for assigning each mixture object in said matrix recursively into one of two defined categories on the basis of presence or absence of a given feature of interest from said set of descriptors and repeating such analysis until each mixture object of said matrix has been classified in terms of presence or absence of given features of interest from said set of descriptors and assigned to a terminal node; and

(f) output means for visually displaying, using computer graphics, the relationships of said descriptors with said mixture objects and classes.

9. A computer software system having a set of instructions for controlling a general purpose digital computer in performing a desired function comprising:

a set of instructions formed into each of a plurality of modules, each module comprising;

(a) an input process responsive to operator commands enabling an operator to specify a set of descriptors and convert said descriptors into a bit string such that each descriptor reflects the presence or absence of a potentially useful feature of interest in a mixture object of interest, wherein each mixture object is a member of a data set that is characterized in being a mixture of classes, each class containing one or more of said mixture objects, and wherein multiple mixture objects present the same trait of interest, but classes of mixture objects produce the response characteristic that is a trait of interest through different underlying mechanisms;

(b) a data storage process for storing the assembled set of (a);

(c) a computational process for executing programmed steps that examine each mixture object of said matrix for presence or absence of each of said descriptors;

(d) a computational process for assembling the results of (c) into a vector for each mixture object and a virtual matrix with each row corresponding to a mixture object and each column corresponding to a feature;

(e) a computational process for assigning each mixture object in said matrix into one of two defined categories on the basis of presence or absence of a given feature of interest from said set of descriptors and repeating such analysis until each member of said matrix has been classified in terms of presence or absence of given features of interest from said set of descriptors and assigned to a terminal node;

(f) a data storage process; and

(g) an output process for visually displaying, using computer graphics, a relationship of said descriptors with said mixture objects and classes.

10. A computer-based method of analyzing biological potency of individual chemical structure features out of a plural mixture of chemical compounds wherein a created data set is characterized in being a mixture of data objects, each data object itself being a mixture of active and/or inactive chemical compounds, which active chemical compounds exhibit a trait of interest, wherein the underlying mechanisms of activity may be through a single or multiple mechanisms, comprising the steps of:

(a) assembling a set of descriptors such that each descriptor captures a chemically useful feature of one or more members of a mixture of chemical compounds such that one member is captured if individual chemical compounds are being decoded, two members are captures if pairs of chemical compounds are being decoded, three members are captured if triples of chemical compounds are being decoded and so on;

(b) examining each member, pair or triple, or so forth, of said mixture of chemical compounds for presence or absence of each of said features of interest;

(c) assembling the results of step (b) into a descriptor vector;

d) comparing the features of the individual compound, pair, triple and so forth, to the features of a terminal node of choice and determining a resident terminal node;

(e) repeating step (d) until each compound, pair, triple and so forth of said set of mixtures of chemical compounds has been identified and characterized in relation to the terminal node it would reside within.

38. A computer-based method of encoding, decoding and identifying individual chemical compounds out of a chemical mixture, comprising the steps of:

(a) assembling the results of previously conducted screening of the chemical mixture for a biological activity of interest;

(b) assembling a set of descriptors such that each descriptor captures a chemically useful feature of one or more members of a chemical mixture;

(c) examining each combination of members of said chemical mixture for presence or absence of each of said descriptors;

(d) correlating presence or absence of said chemical descriptors with an assigned terminal node, thereby identifying predicted activity; and

(e) analyzing subsequent chemical mixtures for chemical structure, comparing their chemical structure against said predicted activity and extrapolating biological reactivity of such subsequent chemical mixtures therefrom.

39. The method as claimed in claim 38, including the additional step of assembling a chemical structure data file.

40. The method as claimed in claim 38, including the additional step of assembling biological data pertaining to each chemical compound and assigning each chemical compound or mixture its biological data.

41. The method as claimed in claim 38, in which said correlation is between presence or absence of one or more chemical descriptors and biological activity of a chemical compound or mixture.

42. The method as claimed in claim 38, in which said correlation is between presence or absence of one or more chemical descriptors and pharmacological activity of a chemical compound or mixture.

43. The method as claimed in claim 38, in which said descriptor is an atom pair.

44. The method as claimed in claim 38, in which said descriptor is an atom triple.

45. The method as claimed in claim 44, in which said atom triple is a set of three defined atoms in a molecule of interest, each atom defined by element, by spatial relation to each of the other two atoms, and by the type of chemical bond or number of chemical bonds separating them in the molecule.

46. The method as claimed in claim 38, in which said descriptor is a molecular fragment.

47. The method as claimed in claim 38, in which said descriptor is a molecular topological torsion.

48. The method as claimed in claim 38, in which said descriptor is a binary of continuous variables.

49. The method as claimed in claim 38, in which said descriptors are a combination in any order of atom pairs, atom triples, molecular fragments, molecular topological torsions, thermodynamic stability descriptors or a binary of continuous variables.

50. The method as claimed in claim 38, in which presence or absence of each feature of intererest is represented as a 1 or a 0, respectively.

51. The method as claimed in claim 38, in which said vector is computationally represented as a bit string.

52. The method as claimed in claim 38, including the additional step of decoding the chemical compounds in said chemical mixture by reference to said matrix vectors for the mixture.

53. The method as claimed in claim 38, in which said recursive partitioning is graphically represented as a recursive partitioning analysis tree.

54. A computer-based method of encoding, identifying and correlating individual genetic features of a genetic polymorphism out of a plural populational mixture of individual subjects so as to identify useful diagnoses and therapies of individuals and in the identification of genes and gene products useful in defining biological targets of interest, comprising the steps of:

(a) assembling a set of descriptors such that each descriptor captures a genetically useful feature, allele, alleles, or marker, of one or more members of a mixture population of individuals having a phenotype of interest;

(b) examining each member of said population of individuals for presence or absence of each of said genetic features;

(c) assembling the results of step (b) into a matrix;

(d) dividing the data in said matrix into one of two defined categories on the basis of presence or absence of a given genetic features from said set of genetic features;

(e) repeating step (d) until each member of said population of individuals has been identified and characterized in terms of presence or absence of each genetic feature; and

(f) correlating presence or absence of said genetic features with known phenotypes of each of said mixture population of individuals, thereby deriving a relationship between genotype and phenotype, said relationship useful in diagnosis and therapy of individuals and also useful for identification of gene products, said gene products useful for selecting drug targets or said gene products useful for determining the genetic origiSn of a disease.

55. The method as claimed in claim 54, including the additional step of assembling a populational phenotype data file.

56. The method as claimed in claim 54, in which said descriptor is an identified allele or marker.

57. The method as claimed in claim 54, in which said descriptor is absence of a given allele or marker.

58. The method as claimed in claim 54, in which each descriptor is an element of a vector in said matrix.

59. The method as claimed in claim 54, in which each individual in said population is encoded by a vector in said matrix.

60. The method as claimed in claim 54, in which presence or absence of each descriptor is represented as a 1 or a 0, respectively.

61. The method as claimed in claim 54, in which said matrix vector is computationally represented as a bit string.

62. The method as claimed in claim 54, in which said bit string is utilized to computationally create a bit string data file.

63. The method as claimed in claim 54, in which said bit string is computationally compressed as a sparse matrix.

64. The method as claimed in claim 54, in which said sparse matrix is statistically analyzed by recursive partitioning.

65. The method as claimed in claim 54, in which said recursive partitioning is performed by the CART method.

66. The method as claimed in claim 54, in which said recursive partitioning is performed by the FIRM method.

67. The method as claimed in claim 54, in which said recursive partitioning is performed by the C4.5 method.

68. The method as claimed in claim 54, in which said FIRM method is converted from multiway splits to binary splits.

69. The method as claimed in claim 54, including the additional step of selecting the descriptor that correlates most closely with the highest average incidence of a phenotype of interest of all individuals in the population that have such a descriptor and creating two subsets of individuals where said descriptor is present or absent, respectively, and repeating this process through subsequent iterations until all descriptors in said descriptor set have been examined and analyzed for prevalence in said population.

70. The method as claimed in claim 54, including the additional step of decoding the individuals in said population by reference to said matrix vectors.

71. The method as claimed in claim 54, in which said recursive partitioning is graphically represented as a recursive partitioning analysis tree.

72. The method as claimed in claim 54, in which said statistical test for splitting a node is a t-test.

73. The method as claimed in claim 54, in which said statistical test for splitting a node is a chi-square test.