Systems and methods for diagnosing a biological specimen using probabilities

• US 7,747,547 B1
• Filed: 02/10/2009
• Issued: 06/29/2010
• Est. Priority Date: 10/31/2007
• Status: Expired due to Fees

• Current Assignee: Cancer Genetics
• Original Assignee: Pathwork Diagnostics, Inc. (Response Genetics, Inc)
• Inventors: Buturovic, Ljubomir J., Anderson, Glenda G.
• Primary Examiner(s): VINCENT, DAVID ROBERT
• Application Number:
  US12/378,165
• Time in Patent Office: 504 Days
• Field of Search: 706/12, 706/45, 706/20, 435/6, 702/19, 702/20, 702/188, 703/11
• US Class Current: 706/12
• CPC Class Codes: G06F 18/2415 based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
• CPC Class Codes: G06V 10/764 using classification, e.g. of video objects
• CPC Class Codes: G06V 2201/04 Recognition of patterns in DNA microarrays
• 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 20/30 Detection of binding sites or motifs
• CPC Class Codes: G16B 25/00 ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
• 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 40/20 Supervised data analysis
• CPC Class Codes: G16B 40/30 Unsupervised data analysis

Abstract

Apparatus, systems and methods for determining, for each respective phenotypic characterization in a set of {T₁, . . . , T_k} characterizations, that a test specimen has the respective characterization are provided. A pairwise probability function g_pq(X, W_pq), for a phenotypic pair (T_p, T_q) in {T₁, . . . , T_k} is learned using a training population. W_pq is a set of parameters derived from Y for (T_p, T_q) by substituting each Y₁ in Y into g_pq(X, W_pq), as X, where Y_i is the set of cellular constituent abundance values from sample i in the training population exhibiting T_p or T_q. The learning step is repeated for each (T_p, T_q) in {T₁ . . . , T_k}, thereby deriving pairwise probability functions G={g_1,2(X, W_1,2), . . . , g_{k-1, k}(X, W_{k-1, k})}. Pairwise probability values P={p_1,2, . . . , p_{k-1, k}} are computed, where each p_pq is equal to g_pq(Z, W_pq) in G, the probability that the test specimen has T_p and not T_q, where Z is cellular constituent abundance values of the test specimen.

Claims

1. A computer implemented method of determining, for each respective phenotypic characterization in a set of {T₁, . . . , T_k} phenotypic characterizations, a probability that a test biological specimen has the respective phenotypic characterization, the method comprising:

(A) learning a pairwise probability function g_pq(X, W_pq) using a training population, for a pair of phenotypic characterizations (T_p, T_q) in the set of {T₁, . . . , T_k} phenotypic characterizations, wherein

(i) there are at least five training samples in the training population for each phenotypic characterization in the set of {T₁, . . . , T_k} phenotypic characterizations;

(ii) Y is the set of all training samples in the training population that exhibits either phenotypic characterization T_p or phenotypic characterization T_q, and each Y_i in Y is the set of {y_i1, . . . , y_in} cellular constituent abundance values for a plurality of cellular constituents measured from a sample i, from the training population, which exhibits either phenotypic characterization T_p or phenotypic characterization T_q;

(iii) W_pq is a set of parameters derived from Y in the learning step (A) for a pair of phenotypic characterizations (T_p, T_q) by substituting each Y_i into g_pq(X, W_pq), as X, during said learning step (A);

(iv) k is 3 or greater;

(v) n is at least 1; and

(vi) p is not equal to q;

(B) repeating the learning step (A) for a different pair of phenotypic characterizations (T_p, T_q), using the training population, for all unique pairs of phenotypic characterizations in the set of {T₁ . . . , T_k} phenotypic characterizations, thereby deriving a plurality of pairwise probability functions G={g_1,2(X, W_1,2), . . . , g_{k-1, k}(X, W_{k-1, k})};

(C) computing a plurality of pairwise probability values P={p_1,2, . . . , p_{k-1, k}}, wherein each pairwise probability value p_pq in P is equal to g_pq(Z, W_pq) in G, the probability that the test biological specimen has phenotypic characterization T_p and does not have phenotypic characterization T_q, wherein Z is a set of {z₁, . . . , z_n} cellular constituent abundance values measured from the test biological specimen for said plurality of cellular constituents;

(D) optionally converting P to a set M of k probabilities, wherein M={p₁, p₂, . . . , p_k}, wherein each probability p_j in M is a probability for a phenotypic characterization in the set of {T₁, . . . , T_k} phenotypic characterizations that the test biological specimen has the phenotypic characterization such that

2. The computer implemented method of claim 1, wherein

3. The computer implemented method of claim 2, wherein s(Z, R_i) is equal to the value of the kernel function e⁽⁻

y[(z<sup>1−

ri1)2+(z2−

ri2)2+ . . . (zn−

rin)2])</sup>, wherein z₁, . . . , z_n are cellular constituent abundance values in Z that respectively correspond to cellular constituent abundance values r_i1, . . . , r_in in R_i; and

wherein s(Z, R_j) is equal to the value of the kernel function e<sup>(−

y[(z1−

rj1)2+(z2−

rj2)2+ . . . (zn−

rjn)2])</sup>, where z₁, . . . , z_n are cellular constituent abundance values in Z that respectively correspond to cellular constituent abundance values r_j1, . . . , r_jn in R_j.

4. The computer implemented method of claim 2, the method further comprising determining values, for the given pair of phenotypic characterization (T_p, T_q), for the set of weights w′

_is, w″

_j, and b used in g_pq(Z, W_pq) before the computing step (C) by subjecting each set of cellular constituent abundance values in the training population that was measured from samples that have phenotypic characterization T_p or T_q to a support vector machine.

5. The computer implemented method of claim 2, wherein

6. The computer implemented method of claim 1, wherein a phenotypic characterization in the plurality of phenotypic characterizations is an organ type, an abnormal state in an organ, an tissue type, an abnormal state in a tissue, a cell type, an abnormal cell type, a cell morphology, an abnormal cell morphology, a disease state, a disease prognosis, or a therapeutic response.

7. The computer implemented method of claim 1, wherein the set of cellular constituent abundance values Z for the plurality of cellular constituents measured from the test biological specimen and the set of cellular constituent abundance values Y_i for the plurality of cellular constituents measured from the sample i from the training population are measured from a microarray comprising probes arranged with a density of 100 different probes per 1 cm² or higher.

8. The computer implemented method of claim 1, wherein the set of cellular constituent abundance values Z for the plurality of cellular constituents measured from the test biological specimen and the set of cellular constituent abundance values Y_i for the plurality of cellular constituents measured from the sample i from the training population are measured from a microarray comprising probes arranged with a density of at least 2,500 different probes per 1 cm².

9. The computer implemented method of claim 1, wherein the set of cellular constituent abundance values Z for the plurality of cellular constituents measured from the test biological specimen and the set of cellular constituent abundance values Y_i for the plurality of cellular constituents measured from the sample i from the training population are measured from a microarray comprising at least 10,000 different probes.

10. The computer implemented method of claim 1, wherein the set of cellular constituent abundance values Z for the plurality of cellular constituents measured from the test biological specimen and the set of cellular constituent abundance values Y_i for the plurality of cellular constituents measured from the sample i from the training population are measured from an expression microarray, a comparative genomic hybridization microarray, an exon microarray, or a microRNA microarray.

11. The computer implemented method of claim 1, wherein the set of cellular constituent abundance values Z for the plurality of cellular constituents measured from the test biological specimen and the set of cellular constituent abundance values Y_i for the plurality of cellular constituents measured from the sample i from the training population are measured from a microarray comprising between 10 oligonucleotides and 5×

10⁶ oligonucleotides.

12. The computer implemented method of claim 1, wherein the plurality of cellular constituents is between 5 mRNAs and 50,000 mRNAs and the cellular constituent abundance values are amounts of each mRNA.

13. The computer implemented method of claim 1, wherein the plurality of cellular constituents is between 50 proteins and 200,000 proteins and the cellular constituent abundance values are amounts of each protein.

14. The computer implemented method of claim 1, wherein, for each respective phenotypic characterization in the plurality of phenotypic characterizations, the training population comprises at least three samples that have the respective phenotypic characterization.

15. The computer implemented method of claim 1, wherein each phenotypic characterization in the plurality of phenotypic characterizations is a cancer tissue of origin and wherein the plurality of phenotypic characterizations comprises bladder cancer, breast cancer, colorectal cancer, gastric cancer, germ cell cancer, kidney cancer, hepatocellular cancer, non-small cell lung cancer, non-Hodgkin'"'"'s lymphoma, melanoma, ovarian cancer, pancreatic cancer, prostate cancer, soft tissue sarcoma, and thyroid cancer.

16. The computer implemented method of claim 1, wherein the plurality of phenotypic characterizations is between 2 phenotypic characterizations and 100 phenotypic characterizations.

17. The computer implemented method of claim 1, further comprising receiving the set of cellular constituent abundance values Z from a remote source over a computer network, and communicating the one or more pairwise probabilities p_pq in P and/or the one or more p_j in M to the remote source over said computer network.

18. The computer implemented method of claim 17, wherein the remote source is a remote computer or a remote computer system.

19. The computer implemented method of claim 1, wherein the converting step (C) comprises deeming the set of probabilities {p₁, p₂, . . . , p_k} that minimize the criterion f(p₁, p₂, . . . , p_k) over p_i, where f( )is defined as

20. The computer implemented method of claim 1, wherein said learning of the pairwise probability function g_pq(X, W_pq) comprises using a decision tree, predictive analysis of microarrays, a multiple additive regression tree, a neural network, a clustering algorithm, principal component analysis, a nearest neighbor analysis, a linear discriminant analysis, a quadratic discriminant analysis, a support vector machine, an evolutionary method, a projection pursuit, a radial basis function, or weighted voting.

21. An apparatus for determining, for each respective phenotypic characterization in a set of {T₁, . . . , T_k} phenotypic characterizations, a probability that a test biological specimen has the respective characterization, the apparatus comprising:

a processor; and

a memory, coupled to the processor, the memory storing a module comprising;

(A) instructions for learning a pairwise probability function g_pq(X, W_pq) using a training population, for a pair of phenotypic characterizations (T_p, T_q) in the set of {T₁, . . . , T_k} phenotypic characterizations, wherein;

(i) there are at least five training samples in the training population for each phenotypic characterization in the set of {T₁, . . . , T_k} phenotypic characterizations;

(ii) Y is the set of all training samples in the training population that exhibits either phenotypic characterization T_p or phenotypic characterization T_q, and each Y_i in Y is the set of {y_i1, . . . , y_in} cellular constituent abundance values for a plurality of cellular constituents measured from a sample i, from the training population, which exhibits either phenotypic characterization T_p or phenotypic characterization T_q;

(iii) W_pq is a set of parameters derived from Y by the instructions for learning (A) for a pair of phenotypic characterizations (T_p, T_q) by substituting each Y_i into g_pq(X, W_pq), as X, during said learning step (A);

(iv) k is 3 or greater;

(v) n is at least 1; and

(vi) p is not equal to q,

(B) instructions for repeating the instructions for learning (A) for a different pair of phenotypic characterizations (T_p, T_q), using the training population, for all unique pairs of phenotypic characterizations in the set of {T₁ . . . , T_k} phenotypic characterizations, thereby deriving a plurality of pairwise probability functions G={g_1,2(X, W_1,2), . . . , g_{k-1, k}(X, W_{k-1, k})};

(C) instructions for computing a plurality of pairwise probability values P={p_1,2, . . . , p_{k-1, k}}, wherein each pairwise probability value p_pq in P is equal to g_pq(Z, W_pq) in G, the probability that the test biological specimen has phenotypic characterization T_p and does not have phenotypic characterization T_q, wherein Z is a set of {z₁, . . . , z_n} cellular constituent abundance values measured from the test biological specimen for said plurality of cellular constituents;

(D) optionally, instructions for converting P to a set M of k probabilities, wherein M={p₁, p₂, . . . , p_k}, wherein each probability p_j in M is a probability for a phenotypic characterization in the set of {T₁, . . . , T_k} phenotypic characterizations that the biological specimen has the phenotypic characterization such that

22. The apparatus of claim 21, said module further comprising instructions for:

receiving the set of cellular constituent abundance values Z from a remote source over a computer network, and

communicating the one or more pairwise probabilities p_pq in P or the one or more p_j in M to the remote source over said computer network.

23. The apparatus of claim 21, wherein the memory further comprises Y and an indication of the phenotypic characterization of each sample i in the training population.

24. The apparatus of claim 21, wherein the network is the Internet.

25. A computer-readable medium storing a computer program executable by a computer to determine, for each respective phenotypic characterization in a set of {T₁, . . . , T_k} phenotypic characterizations, a probability that a test biological specimen has the respective phenotypic characterization, the computer program comprising:

(A) instructions for learning a pairwise probability function g_pq(X, W_pq) using a training population, for a pair of phenotypic characterizations (T_p, T_q) in the set of {T₁, . . . , T_k} phenotypic characterizations, wherein;

(i) there are at least five training samples in the training population for each phenotypic characterization in the set of {T₁, . . . , T_k} phenotypic characterizations;

(ii) Y is the set of all training samples in the training population that exhibits either phenotypic characterization T_p or phenotypic characterization T_q, and each Y_i in Y is the set of {y_i1, . . . , y_in} cellular constituent abundance values for a plurality of cellular constituents measured from a sample i, from the training population, which exhibits either phenotypic characterization T_p or phenotypic characterization T_q;

(iii) W_pq is a set of parameters derived from Y in the learning step (A) for a pair of phenotypic characterizations (T_p, T_q) by substituting each Y_i into g_pq(X, W_pq), as X, by the instructions for learning (A);

(iv) k is 3 or greater;

(v) n is at least 1; and

(vi) p is not equal to q,

(B) instructions for repeating the instructions for learning (A) for a different pair of phenotypic characterizations (T_p, T_q), using the training population, for all unique pairs of phenotypic characterizations in the set of {T₁ . . . , T_k} phenotypic characterizations, thereby deriving a plurality of pairwise probability functions G={g_1,2(X, W_1,2), . . . , g_{k-1, k}(X, W_{k-1, k})};

(C) instructions for computing a plurality of pairwise probability values P={p_1,2, . . . , p_{k-1, k}}, wherein each pairwise probability value p_pq in P is equal to g_pq(Z, W_pq) in G, the probability that the test biological specimen has phenotypic characterization T_p and does not have phenotypic characterization T_q, wherein Z is a set of {z₁, . . . , z_n} cellular constituent abundance values measured from the test biological specimen for said plurality of cellular constituents;

(D) optionally, instructions for converting P to a set M of k probabilities, wherein M={p₁, p₂, . . . , p_k}, wherein each probability p_j in M is a probability for a phenotypic characterization in the set of {T₁, . . . , T_k} phenotypic characterizations that the biological specimen has the phenotypic characterization such that