Method for determining biological expression levels by linear programming

• US 7,031,845 B2
• Filed: 07/19/2002
• Issued: 04/18/2006
• Est. Priority Date: 07/19/2002
• Status: Expired due to Fees

• Current Assignee: University of Chicago
• Original Assignee: University of Chicago
• Inventors: Wright, Stephen J., Clark, Terry, Kurtz, Stuart A., Dyanov, Chris (Hristem), Scott, Ridgway, Quigg, Richard
• Primary Examiner(s): Moran, Marjorie A.
• Assistant Examiner(s): MILLER, MARINA I
• Application Number:
  US10/198,141
• Publication Number:
  US 20040014044A1
• Time in Patent Office: 1,369 Days
• Field of Search: 435/6
• 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

Abstract

A method for determining a matrix of expression levels corresponding to a set of biological targets (e.g., genes or gene fragments) and a set of biological samples, including obtaining a matrix of signal values corresponding to the set of biological targets; computing a vector of expression levels for a sample in the set of biological samples using the matrix of signal values; storing the vector of computed expression levels in a storage matrix; repeating the computing and storing steps for each sample in the set of biological samples; and outputting the storage matrix as the matrix of expression levels. The method, based on a linear programming formulation of the problem, works for both “promiscuous” probe array data, in which there may be multiple targets indicated by a single probe, and the “polygamous” case, in which there are multiple probes for a single target. The preferred method can also process data obtained from multiple SAGE analyses using multiple markers. A second embodiment of the method determines optimal expression levels when the available probe data is noisy or uncertain.

Claims

1. A computer-implemented method for determining, for a biological sample, a vector of expression levels, each expression level representing a quantity of a target, of a corresponding set of biological targets, that is present in the biological sample, the method comprising:

obtaining a matrix of signal values P corresponding to the set of biological targets;

obtaining a vector of signal values A corresponding to the biological sample;

determining a vector E, a vector s, and a vector t that minimize a sum of all elements of s and t, and satisfy a constraint PE+s−

t=A, wherein the elements of the vectors E, s, and t are nonnegative real numbers; and

outputting the vector E determined in the determining step as the vector of expression levels.

2. The method of claim 1, wherein obtaining the vector of signal values A comprises:

testing the sample with an array of reactive probes.

3. The method of claim 1, wherein obtaining the vector of signal values A comprises:

testing the sample with an array of reactive probes,

at least one probe in the array of reactive probes being indicative of more than one target in the set of biological targets.

4. The method of claim 1, wherein obtaining the vector of signal values A comprises:

testing the sample with an array of reactive probes,

each probe in the array of reactive probes being a sequence of oligonucleotides.

5. The method of claim 1, wherein obtaining the vector of signal values A comprises:

testing the sample with an array of reactive probes including at least one probe which is a sequence of oligonucleotides from at least one part of a known gene.

6. The method of claim 1, wherein the determining step comprises:

identifying a set of biological targets S(A) consistent with the vector of signal values A; and

setting, for each target in the set of biological targets that is not in S(A), a respective element of the vector E to zero.

7. The method of claim 1, wherein the determining step comprises:

using a linear programming algorithm to determine the vector E, the vector s, and the vector t.

8. The method of claim 7, wherein the determining step further comprises:

constructing a feasible starting point for the linear programming algorithm, for an initial nonnegative vector E_o, by initializing an l-th element of the vector s to be s_l=−

min((PE_o−

A)_l,0), and initializing an i-th element of the vector t to be t_l=−

max((PF_o−

A)_l,0).

9. The method of claim 8, wherein the determining step further comprises:

constructing the initial nonnegative vector E_o by setting each element of E_o to be either zero or, if a corresponding biological target could have uniquely contributed to the vector of signal values A, to a predetermined positive value.

10. The method of claim 1, wherein obtaining the vector of signal values A comprises:

generating a plurality of short nucleotide sequence tags from the sample;

concatenating the plurality of short nucleotide sequence tags into a single molecule;

sequencing the molecule to determine a count for each tag in the plurality of short nucleotide sequence tags; and

mapping the counts determined in the sequencing step into the vector of signal values A.

11. The method of claim 1, wherein the matrix of signal values P obtained in the obtaining step includes more columns than rows.

12. The method of claim 1, wherein obtaining the matrix of signal values P comprises:

testing each target in the set of biological targets with an array of reactive probes.

13. The method of claim 1, wherein obtaining the matrix of signal values P comprises:

testing each target in the set of biological targets with the array of reactive probes, each target being a gene or a gene fragment.

14. The method of claim 1, wherein obtaining the matrix of signal values P comprises:

generating a plurality of short nucleotide sequence tags for a target in the set of biological targets;

concatenating the plurality of short nucleotide sequence tags into a single molecule;

sequencing the molecule to determine a count for each tag in the plurality of short nucleotide sequence tags;

mapping the counts determined in the sequencing step into the matrix of signal values P; and

repeating the generating, concatenating, sequencing, and mapping steps for each target in the set of biological targets.

15. The method of claim 1, further comprising:

storing the vector of expression levels determined in the determining step in a storage matrix;

repeating the step of obtaining the vector of signal values A, the determining step, and the storing step for at least one biological sample; and

outputting the storage matrix as a matrix of expression levels.

16. A computer-implemented method for determining, for a biological sample, a vector of expression levels, each expression level representing a quantity of a target, of a corresponding set of biological targets, that is present in the biological sample, the method comprising:

obtaining a matrix of signal values P corresponding to the set of biological targets;

obtaining a vector of lower signal values L and a vector of higher signal values H corresponding to the sample, each element of L being less than or equal to a respective element of H;

determining a vector E, a vector s, and a vector t, that minimize a total sum of all elements of s and t, and satisfy constraints s≧

L−

PE and t≧

PE−

H, wherein the elements of the vectors E, s, and t are nonnegative real numbers; and

outputting the vector E determined in the determining step as the vector of expression levels.

17. A computer program product comprising a computer storage medium configured to store plural computer program instructions which, when executed by a computer, causes the computer to determine, for a biological sample, a vector of expression levels, each expression level representing a quantity of a target, of a corresponding set of biological targets, that is present in the biological sample, by performing plural steps comprising:

obtaining a matrix of signal values P corresponding to the set of biological targets;

obtaining a vector of signal values A corresponding to the biological sample;

determining a vector E, a vector s, and a vector t that minimize a total sum of all elements of s and t, and satisfy a constraint PE+s−

t=A, wherein the elements of the vectors E, s, and t are nonnegative real numbers; and

outputting the vector E determined in the determining step as the vector of expression levels.