T-cell epitope prediction

• US 8,121,797 B2
• Filed: 12/21/2007
• Issued: 02/21/2012
• Est. Priority Date: 01/12/2007
• Status: Active Grant

• Current Assignee: Zhigu Holdings Limited
• Original Assignee: Microsoft Corporation
• Inventors: Heckerman, David E., Kadie, Carl M., Listgarten, Jennifer, Zaitlen, Noah Aaron, Jojic, Nebojsa
• Primary Examiner(s): Skowronek, Karlheinz R
• Application Number:
  US11/963,081
• Publication Number:
  US 20080172215A1
• Time in Patent Office: 1,523 Days
• Field of Search: None
• US Class Current: 702/20
• CPC Class Codes: G16B 15/00 ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
• CPC Class Codes: G16B 15/20 Protein or domain folding
• 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 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

Abstract

Epitope prediction models are described herein. By way of example, a system for predicting epitope information relating to a epitope can include a classification model (e.g., logistic regression model). The trained classification model can illustratively operatively execute one ore logistic functions on received protein data, and incorporate one or more of hidden binary variables and shift variables that when processed represent the identification (e.g., prediction) of one or more desired epitopes. The classification model can be configured to predict the epitope information by processing data including various features of an epitope, MHC, MHC supertype, and Boolean combinations thereof.

Claims

1. A computer implemented method that facilitates epitope prediction, comprising:

training, using a processing unit, a logistic regression (LR) model for epitope prediction using information from a plurality of sources representative of standard and special features of a desired epitope,

wherein the standard features comprise, alone or in conjunction, data representative of an identity and/or supertype of a major histocompatibility complex (MHC) allele, and data representative of the identity and/or a chemical property of an amino acid at a certain position of an epitope,

wherein the special features comprise, alone or Boolean combinations of, data representative of the standard features, the identity of an amino acid and/or the chemical property of the amino acid at a given position along either region that flanks an epitope and data representative of an amino acid and/or the chemical property of the amino acid at a given position along a MHC molecule;

employing hidden variables that represent an absence of supertypes among MHC molecules;

employing a shift variable that represents a position of a peptide within a groove of the MHC molecule; and

performing a multi-factor cross validation to confirm the epitope prediction.

2. The method as recited in claim 1, further comprising training the LR model on epitopes within HLA alleles and HLA supertypes.

3. A computer readable storage medium having computer readable instructions to instruct a computer to perform a method comprising:

training a logistic regression (LR) model for epitope prediction using information from a plurality of sources representative of standard and special features of a desired epitope,

wherein the standard features comprise, alone or in conjunction, data representative of an identity and/or supertype of a major histocompatibility complex (MHC) allele, and data representative of the identity and/or a chemical property of an amino acid at a certain position of an epitope,

wherein the special features comprise, alone or Boolean combinations of, data representative of the standard features, the identity of an amino acid and/or the chemical property of the amino acid at a given position along either region that flanks an epitope and data representative of an amino acid and/or the chemical property of the amino acid at a given position along a MHC molecule;

using hidden variables that represent a presence of supertypes among MHC molecules;

employing a shift variable that represents a position of a peptide within a groove of the MHC molecule; and

performing a multi-factor cross validation to confirm the epitope prediction.

4. A computer implemented method that facilitates epitope prediction, comprising:

training, using a processing unit, a logistic regression (LR) model for epitope prediction using one or more hidden variables representative of one or more characteristics of a major histocompatibility complex (MHC) molecule;

using special features that include an identity of an amino acid and/or a chemical property of the amino acid at a given position along either region that flanks an epitope and the identity of the amino acid and/or the chemical property of the amino acid at a given position along the MHC molecule; and

performing a multi-factor cross validation to confirm the epitope prediction.

5. The method as recited in claim 4, further comprising identifying the one or more hidden variables representative of a presence of one or more supertypes among MHC molecules.

6. The method as recited in claim 5, further comprising identifying the one or more hidden variables representative of an absence of one or more supertypes among MHC molecules.

7. The method as recited in claim 4, further comprising training the LR model using an expectation-maximization (EM) algorithm wherein a maximization calculus comprises a gradient optimization.

8. The method as recited in claim 7, further comprising determining a probability whether an observed peptide-MHC pair is an epitope using a standard inference.

9. A computer implemented method that facilitates epitope prediction, comprising:

training, using a processing unit, a logistic regression (LR) model for epitope prediction using one or more shift variables representative of a position of a peptide within a groove of a major histocompatibility complex (MHC) molecule;

using data representative of an identity of a major histocompatibility complex (MHC) allele and data representative of the identity of an amino acid at a certain position of an epitope; and

performing a multi-factor cross validation to confirm the epitope prediction.

10. The method as recited in claim 9, further comprising adding a shift variable to the LR model representative of a single position of a peptide within the groove of the MHC molecule.

11. The method as recited in claim 9, further comprising conditioning variables of the LR model on a value of the one or more shift variables.

12. The method as recited in claim 9, further comprising training the LR model using an expectation-maximization (EM) algorithm wherein a maximization calculus comprises a gradient optimization.

13. The method as recited in claim 12, further comprising determining a probability whether an observed peptide-MHC pair is an epitope using a standard interference.

14. A system that facilitates predicting an epitope, the system stored on computer-readable storage media, the system comprising:

a prediction component configured to predict epitope information by employing information from a plurality of sources including human leukocyte antigen (HLA) alleles and HLA supertypes to be processed by a classification model;

a classification model engine executing a selected trained classification model employing information using standard and special features of the epitope,

wherein the trained classification model is trained to include one or more hidden binary variables that represent a presence or an absence of the HLA supertypes,

wherein the trained classification model is trained to include one or more shift variables to generate the epitope prediction,

wherein the trained classification model is trained within the HLA alleles and within the HLA supertypes to generate the epitope prediction, and

the classification model engine further executing the selected trained classification model to perform an optimization by determining a global maximum; and

the classification model engine further executing the selected trained classification model to perform a multi-factor cross validation to confirm the epitope prediction.

15. The system as recited in claim 14, wherein the classification model comprises a neural network having artificial intelligence capabilities.

16. The system as recited in claim 14, wherein the classification model comprises a logistic regression (LR) model.

17. The system as recited in claim 16, wherein the LR model is given by:

18. The system as recited in claim 17, use a data set of cases (y¹,x¹), . . . , (yⁿ,x^m) that are identically distributed given the model parameters, the weights are mutually independent, each having a Gaussian prior p(w_i|ρ

²)=N(0,ρ

²), wherein determining the weights that have the maximum a posteriori (MAP) probability are determined by the quantity;

19. The system as recited in claim 14, wherein the classification model employs an expectation-maximization (EM) algorithm and a standard inference in determining a probability of whether a given peptide-MHC pair is an epitope.

20. The system as recited in claim 19, wherein a maximization calculus of the EM algorithm comprises a gradient optimization.