Methods of characterizing drug activities using consensus profiles

• US 6,801,859 B1
• Filed: 12/23/1998
• Issued: 10/05/2004
• Est. Priority Date: 12/23/1998
• Status: Expired due to Term

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Rosetta Inpharmatics LLC (Merck & Co., Inc.)
• Inventors: He, Yudong, Friend, Stephen H., Stoughton, Roland
• Primary Examiner(s): Marschel, Ardin H.
• Application Number:
  US09/220,142
• Time in Patent Office: 2,113 Days
• Field of Search: 435/6, 435/7.1, 435/7.2, 435/69.1, 702/19, 703/11, 436/501
• US Class Current: 702/19
• CPC Class Codes: G16B 25/00 ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
• CPC Class Codes: G16B 25/10 Gene or protein expression profiling; Expression-ratio estimation or normalisation
• 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 40/30 Unsupervised data analysis

Abstract

The present invention provides methods for enhanced detection of biological response profiles. In particular, the methods of this invention allow for the detection of biological response patterns, such as gene expression patterns, in response to different drug treatments. The methods of the invention also allow the determination of a “consensus profile” which describes a particular class or type of biological response. In certain embodiments the consensus profile may describe the biological response of a particular group or class of drugs. In other embodiments, the consensus profile may describe an “ideal” biological response such as one associated with a desired therapeutic effect. The methods of the present invention also allow for the comparison of different biological responses. Thus, the methods of the invention may be used, e.g., to identify and/or study new drugs.

Claims

1. A method of determining a consensus profile for a first plurality of drug perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of cellular constituents in a plurality of response profiles, whether said set of cellular constituents is upregulated or downregulate by each of said first plurality of drug perturbations, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different drug perturbation among said first plurality of drug perturbations to said type of cell or organism, wherein each set of cellular constituents in said plurality of sets of cellular constituents consists of cellular constituents that co-vary in the plurality of response profiles, wherein the cellular constituents which co-vary are identified by cluster analysis of cellular constituents in the plurality of response profiles, wherein the cluster analysis is done by means of the clustering algorithm hclust, wherein said plurality of response profiles comprises at least five response profiles, and wherein said consensus profile for said first plurality of drug perturbations consists of measurements of said set or sets of cellular constituents that are determined in said determining step to be upregulated or downregulate by each of said first plurality of drug perturbations.

2. A method of determining a consensus profile for a first plurality of drug perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of cellular constituents in a plurality of response profiles, whether said set of cellular constituents is upregulated or downregulated by each of said first plurality of drug perturbations, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different drug perturbation among said first plurality of drug perturbations to said type of cell or organism, wherein each set of cellular constituents in said plurality of sets of cellular constituents consists of cellular constituents that co-vary in the plurality of response profiles, wherein the cellular constituents which co-vary are identified by cluster analysis of cellular constituents in the plurality of response profiles, wherein said plurality of response profiles comprises at least five response profiles, wherein said cluster analysis is carried out by a hierarchical clustering method, wherein said consensus profile for said first plurality of drug perturbations consists of measurements of said set or sets of cellular constituents that are determined in said determining step to be upregulated or downregulated by each of said first plurality of drug perturbations, wherein a statistical significance for the sets of co-varying cellular constituents is determined by means of an objective statistical test, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of co-varying cellular constituents an actual fractional improvement in the cluster analysis of the cellular constituents based on the unpermuted responses of said cellular constituents, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster;

(b) generating permuted responses of cellular constituents by means of Monte Carlo randomization of perturbation index for the response of each cellular constituent across all perturbations;

(c) performing said cluster analysis on the permuted responses of cellular constituents;

(d) determining for each cluster which is generated in step (c) and defines a set of co-varying cellular constituents the fractional improvement in the cluster analysis of cellular constituents based on the permuted responses of cellular constituents, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster; and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the cellular constituents is obtained for each said cluster which is generated by said cluster analysis and defines a set of co-varying cellular constituents;

wherein the statistical significance of each of said sets of co-varying cellular constituents is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.

3. A method of determining a consensus profile for a first plurality of drug perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of cellular constituents in a plurality of response profiles, whether said set of cellular constituents is upregulated or downregulated by each of said first plurality of drug perturbations, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different drug perturbation among said first plurality of drug perturbations to said type of cell or organisms, wherein each set of cellular constituents in said plurality of sets of cellular constituents consists of cellular constituents that co-vary under a second plurality of perturbations or that are co-regulated, wherein said plurality of response profiles comprises at least five response profiles, and wherein said consensus profile for said first plurality of drug perturbations consists of measurements of said set or sets of cellular constituents that are determined in said determining step to be upregulated or downregulated by each of said first plurality of drug perturbations;

wherein the one or more sets of cellular constituents are identified by re-ordering the response profiles into sets associated with similar biological effects, wherein the sets of response profiles associated with similar biological effects are identified by cluster analysis of the response profiles done by means of the clustering algorithm hclust.

4. A method of determining a consensus profile for a first plurality of drug perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of cellular constituents in a plurality of response profiles, whether said set of cellular constituents is unregulated or downregulate by each of said first plurality of drug perturbations, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different drug perturbation among said first plurality of drug perturbations to said type of cell or organism, wherein each set of cellular constituents in said plurality of sets of cellular constituents consists of cellular constituents that co-vary under a second plurality of perturbations or that are co-regulated, wherein said plurality of response profiles comprises at least five response profiles, and wherein said consensus profile for said first plurality of drug perturbations consists of measurements of said set or sets of cellular constituents that are determined in said determining step to be upregulated or downregulated by each of said first plurality of drug perturbations,

wherein the one or more sets of cellular constituents are identified by re-ordering the response profiles into sets associated with similar biological effects, wherein the sets of response profiles associated with similar biological effects are identified by cluster analysis of the response profiles, wherein said cluster analysis is carried out by a hierarchical clustering method, wherein a statistical significance for the sets of response profiles is determined by means of an objective statistical test, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of response profiles an actual fractional improvement in the cluster analysis of the unpermuted response profiles, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster;

(b) generating permuted response profiles by means of Monte Carlo randomization of cellular constituent index for each response profile across the measured cellular constituents;

(c) performing said cluster analysis on the permuted response profiles;

(d) determining for each cluster which is generated in step (c) and defines a set of response profiles the fractional improvement in the cluster analysis of the permuted response profiles, wherein said factional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster; and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the response profiles is obtained for each said cluster which is generated by said cluster analysis and defines a set of response profiles;

wherein the statistical significance of each of said sets of response profiles is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.

5. A method for grouping measured response profiles in sets which are associated with similar biological effects comprising grouping response profiles among a plurality of response profiles into sets, each of said sets of response profiles consisting of response profiles in which the responses of one or more sets of genes in each response profile are similar among response profiles in the set, each response profile in said plurality of response profiles (i) comprising measurements of transcript levels of a plurality of genes, and (ii) resulting from a different perturbation, wherein each of said sets of genes consists of genes that co-vary under a plurality of perturbations or that are co-regulated, wherein said plurality of response profiles comprises at least five response profiles, wherein the sets of response profiles are identified by cluster analysis of the response profiles done by means of the clustering algorithm hclust.

6. A method for grouping measured response profiles in sets which are associated with similar biological effects comprising grouping response profiles among a plurality of response profile into sets, each of said sets of response profiles consisting of response profiles in which the responses of one or more sets of genes in each response profile are similar among response profiles in the set, each response profile in said plurality of response profiles (i) comprising measurements of transcript levels of a plurality of genes, and (ii) resulting from a different perturbation, wherein each of said sets of genes consists of genes that co-vary under a plurality of perturbations or that are co-regulated, wherein said plurality of response profiles comprises at least five response profiles, wherein the sets of response profiles are identified by cluster analysis of the response profiles, wherein said cluster analysis is carried out by a hierarchical clustering method, wherein a statistical significance for the sets of response profiles is determined by means of an objective statistical test, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of response profiles an actual fractional improvement in the cluster analysis of the unpermuted response profiles, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster,

(b) generating permuted response profiles by means of Monte Carlo randomization of gene index for each response profile across the measured genes;

(c) performing said cluster analysis on the permuted response profiles;

(d) determining for each cluster which is generated in step (c) and defines a set of response profiles the fractional improvement in the cluster analysis of the permuted response profiles, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster; and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the response profiles is obtained for each cluster which is generated by said cluster analysis and defines a set of response profiles;

wherein the statistical significance of each of said sets of response profiles is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.

7. A method for determining the therapeutic efficacy of a drug or drug candidate comprising identifying one or more groups of sets of cellular constituents in one or more response profiles associated with exposure to the drug or drug candidate, each response profile comprising measurements of a plurality of cellular constituents, wherein each of said groups is indicative of a particular therapeutic effect, and wherein the therapeutic effect of the drug or drug candidate is determined to be the particular therapeutic effect indicated by the identified groups, wherein each of said sets of cellular constituents consists of cellular constituents that co-vary under a plurality of perturbations or that are co-regulated, wherein the sets of cellular constituents are determined by a method comprising performing cluster analysis of the response profiles done by means of the clustering algorithm hclust.

8. A method for determining the therapeutic efficacy of a drug or drug candidate comprising identifying one or more groups of sets of cellular constituents in one or more response profiles associated with exposure to the drug or drug candidate, each response profile comprising measurements of a plurality of cellular constituents, wherein each of said groups is indicative of a particular therapeutic effect, and wherein the therapeutic effect of the drug or drug candidate is determined to be the particular therapeutic effect indicated by the identified groups, wherein each of said sets of cellular constituents consists of cellular constituents that co-vary under a plurality of perturbations or that are co-regulated, wherein the sets of cellular constituents are determined by a method comprising performing cluster analysis of the response profiles, wherein said cluster analysis is carried out by a hierarchical clustering method, wherein a statistical significance for the sets of cellular constituents is determined by means of an objective statistical test, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of cellular constituents an actual fractional improvement in the cluster analysis of cellular constituents based on the unpermuted responses of said cellular constituents, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster;

(b) generating permuted responses of cellular constituents by means of Monte Carlo randomization of the perturbation index for each cellular constituent across all perturbations;

(c) performing said cluster analysis on the permuted responses of cellular constituents;

(d) determining for each cluster which is generated in step (c) and defines a set of cellular constituents the fractional improvement in the cluster analysis of cellular constituents based on the permuted responses of cellular constituents, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster, and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the cellular constituents is obtained for each cluster which is generated by said cluster analysis and defines a set of cellular constituents;

wherein the statistical significance of each of said sets of cellular constituents is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.

9. A method of determining a consensus profile for a first plurality of drug perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of cellular constituents in a plurality of response profiles, whether said set of cellular constituents is unregulated or downregulated by each of said first plurality of drug perturbations, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different drug perturbation among said first plurality of drug perturbations to said type of cell or organism, wherein each set of cellular constituents in said plurality of sets of cellular constituents consists of cellular constituents that co-vary under a second plurality of perturbations or that are co-regulated, wherein said plurality of response profiles comprises at least five response profiles, and wherein said consensus profile for said first plurality of drug perturbations consists of measurements of said set or sets of cellular constituents that are determined in said determining step to be upregulated or downregulated by each of said first plurality of drug perturbations,

wherein said sets of cellular constituents are co-varying cellular constituent sets that are identified by cluster analysis done by means of the clustering algorithm hclust.

10. A method of determining a consensus profile for a first plurality of drug perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of cellular constituents in a plurality of response profiles, whether said set of cellular constituents is unregulated or downregulated by each of said first plurality of drug perturbations, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different drug perturbation among said first plurality of drug perturbations to said type of cell or organism, wherein each set of cellular constituents in said plurality of sets of cellular constituents consists of cellular constituents that co-vary under a second plurality of perturbations or that are co-regulated, wherein said plurality of response profiles comprises at least five response profiles, and wherein said consensus profile for said first plurality of drug perturbations consists of measurements of said set or sets of cellular constituents that are determined in said determining step to be unregulated or downregulated by each of said first plurality of drug perturbations,

wherein said sets of cellular constituents are co-varying cellular constituent sets that are identified by cluster analysis, wherein said cluster analysis is carried out by a hierarchical clustering method, wherein a statistical significance for the sets of co-varying cellular constituents is determined by means of an objective statistical test, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of co-varying cellular constituents an actual fractional improvement in the cluster analysis of the cellular constituents based on the unpermuted responses of said cellular constituents, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster;

(b) generating permuted responses of cellular constituents by means of Monte Carlo randomization of the perturbation index for response of each cellular constituent across the set of perturbations;

(c) performing said cluster analysis on the permuted responses of cellular constituents;

(d) determining for each cluster which is generated in step (c) and defines a set of co-varying cellular constituents the fractional improvement in the cluster analysis of cellular constituents based on the permuted response responses of cellular constituents, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster; and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the cellular constituents is obtained for each cluster which is generated by said cluster analysis and defines a set of co-varying cellular constituents,

wherein the statistical significance of each of said sets of co-varying cellular constituents is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.

11. A method for grouping measured response profiles in sets which are associated with similar biological effects comprising grouping response profiles in sets among a plurality of response profiles by cluster analysis of said plurality of response profiles, said sets of response profiles consisting of response profiles having similar responses of a group of cellular constituents, each response profile in said plurality of response profiles (i) comprising measurements of a plurality of cellular constituents, and (ii) resulting from a different perturbation, wherein a statistical significance for the sets of response profiles is determined by means of an objective statistical test, wherein said cluster analysis is carried out by a hierarchical clustering method, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of response profiles an actual fractional improvement in the cluster analysis of the unpermuted response profiles, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster;

(b) generating permuted response profiles by means of Monte Carlo randomization of cellular constituent index for each response profile across the measured cellular constituents;

(c) performing said cluster analysis on the permuted response profiles;

(d) determining for each cluster which is generated in step (c) and defines a set of response profiles the fractional improvement in the cluster analysis of the permuted response profiles, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster; and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the response profiles is obtained for each cluster which is generated by said cluster analysis and defines a set of response profiles;

wherein the statistical significance of each of said sets of response profiles is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.

12. A method of determining a consensus profile for a first plurality of perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of genes in a plurality of response profiles, whether said set of genes is upregulated or downregulated by each of said first plurality of perturbations, each response profile in said plurality of response profiles (i) comprising measurements of transcript levels for a plurality of genes, and (ii) resulting from a different perturbation among said first plurality of perturbations to said type of cell or organism, wherein each set of genes in said plurality of sets of genes consists of genes having transcripts that co-vary under a second plurality of perturbations or that are co-regulated, and wherein said consensus profile for said first plurality of perturbations consists of measurements of transcript levels for said set or sets of genes that are determined in said determining step to be upregulated or downregulated by each of said first plurality of perturbations,

wherein each of the sets of genes consists of genes which co-vary in the plurality of response profiles, wherein the genes which co-vary are identified by cluster analysis of genes in the plurality of response profiles, and wherein the cluster analysis is done by means of the clustering algorithm hclust.

13. A method of determining a consensus profile for a firs plurality of perturbations to a cell type or organism, said method comprising determining, for each of a plurality of sets of genes in a plurality of response profiles, whether said set of genes is upregulated or downregulated by each of sad first plurality of perturbations, each response profile in said plurality of response profiles (i) comprising measurements of transcript levels for a plurality of genes, and (ii) resulting from a different perturbation among said first plurality of perturbations to said type of cell or organism, wherein each set of genes in said plurality of sets of genes consists of genes having transcripts that co-vary under a second plurality of perturbations or that are co-regulated, and wherein said consensus profile for said first plurality of perturbations consists of measurements of transcript levels for sad set or sets of genes that are determined in said determining step to be upregulated or downregulated by each of sad first plurality of perturbations, p1 wherein each of the sets of genes consists of genes which co-vary in the plurality of response profiles, wherein the genes which co-vary are identified by cluster analysis of genes in the plurality of response profiles, wherein said cluster analysis is carried out by a hierarchical clustering method, wherein a statistical significance for the sets of co-varying genes is determined by means of an objective statistical test, and wherein the objective statistical test comprises:

(a) determining for each cluster which is generated by said cluster analysis and defines a set of co-varying genes an actual fractional improvement in the cluster analysis of the genes based on the unpermuted responses of said genes, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster;

(b) generating permuted responses of genes by means of Monte Carlo randomization of perturbation index for the response of each gene across all perturbations;

(c) performing cluster analysis on the permuted responses of genes;

(d) determining for each cluster which is generated in step (c) and defines a set of co-varying genes the fractional improvement in the cluster analysis of genes based on the permuted responses of genes, wherein said fractional improvement is an improvement in total scatter with respect to the center of said cluster as compared to total scatter with respect to the respective centers of the two clusters branching out of said cluster; and

(e) repeating steps (b) through (d) so that a distribution of fractional improvements in the cluster analysis of the genes is obtained for each cluster which is generated by said cluster analysis and defines a set of co-varying genes;

wherein the statistical significance of each of said sets of co-varying genes is determined by comparing the actual fractional improvement for the cluster defining said set to the distribution of fractional improvements for the cluster defining said set.