System and method for improving clinical decisions by aggregating, validating and analysing genetic and phenotypic data

• US 20060052945A1
• Filed: 12/03/2004
• Published: 03/09/2006
• Est. Priority Date: 09/07/2004
• Status: Active Grant

• Current Assignee: Natera Incorporated (Natera)
• Original Assignee: Gene Security Network Inc
• Inventors: Rabinowitz, Matthew, Croswell, John, Sarbaev, Miro, Banjevic, Milena, Chambliss, Wayne
• Granted Patent: US 8,024,128 B2
• Time in Patent Office: Days
• US Class Current: 702/20
• 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: G16H 10/60 for patient-specific data, e.g. for electronic patient records
• CPC Class Codes: G16H 50/20 for computer-aided diagnosis, e.g. based on medical expert systems
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

The information management system disclosed enables caregivers to make better decisions, faster, using aggregated genetic and phenotypic data. The system enables the integration, validation and analysis of genetic, phenotypic and clinical data from multiple subjects who may be at distributed facilities. A standardized data model stores a range of patient data in standardized data classes that encompass patient profile information, patient symptomatic information, patient treatment information, and patient diagnostic information including genetic information. Data from other systems is converted into the format of the standardized data classes using a data parser, or cartridge, specifically tailored to the source system. Relationships exist between standardized data classes that are based on expert rules and statistical models. The relationships are used both to validate new data, and to predict phenotypic outcomes based on available data. The prediction may relate to a clinical outcome in response to a proposed intervention by a caregiver. The statistical models may be inhaled into the system from electronic publications that define statistical models and methods for training those models, according to a standardized template. Methods are described for selecting, creating and training the statistical models to operate on genetic, phenotypic and clinical data, in particular for underdetermined data sets that are typical of genetic information. The disclosure also describes how security of the data is maintained by means of a robust security architecture, and robust user authentication such as biometric authentication, combined with application-level and data-level access privileges.

Claims

1. A method for predicting a first phenotypic outcome for a first subject, based on a first set of genetic, phenotypic or clinical data from the first subject, and a second set of genetic, phenotypic or clinical data from a group of second subjects for whom the first phenotypic outcomes are known, the method comprising:

structuring the second set of data from the group of second subjects according to a first set of data classes that have unambiguous definition and encompass all the relevant genetic, phenotypic or clinical data, termed the first set of standardized data classes;

structuring the first set of data from the first subject according to the first set of standardized data classes;

creating a first statistical model for predicting the feature corresponding to the first phenotypic outcome based on the first set of standardized data classes;

training the first statistical model based on the second set of data from the group of second subjects together with their measured first phenotypic outcome;

applying the trained first statistical model to the first set of data to predict the first phenotypic outcome for the first subject.

2. A method according to claim 1 wherein the first phenotypic outcome is the clinical outcome of a human patient in response to a first intervention proposed by a caregiver, and the second set of subjects includes human patients who received the first intervention and for whom the clinical outcome has been determined.

3. A method according to claim 1 wherein the first phenotypic outcome is a clinical response of the first subject to a proposed first intervention, and a first intervention is automatically selected based on optimizing the expected clinical response for the first subject.

4. A method according to claim 1 wherein stored relationships between the standardized data classes, in the form of a first set of expert rules and a second set of statistical models, are used to validate one or more of:

the first set of genetic, phenotypic or clinical data for the first subject;

the second set of genetic, phenotypic or clinical data for the second subjects;

the prediction of the first phenotypic outcome based on the first statistical model;

treatment recommendations based on the prediction of the first phenotypic outcome.

5. A method according to claim 1 wherein the first phenotypic outcome is predicted together with a confidence estimate of the accuracy of that prediction.

6. A method according to claim 1 wherein the first statistical model is automatically selected from multiple relevant statistical models in order to make the most accurate prediction of the first phenotype, based on the second set of data available to train the statistical model, and the first set of data available to make the first prediction for the first subject.

7. A method according to claim 1 wherein the first or second set of genetic and phenotypic data is derived from multiple different external database systems, and is inhaled and formatted according to the first set of standardized data classes by means of a cartridge designed to parse the data from each external system.

8. A method according to claim 1 wherein the first standardized data classes are based on a 20 set of existing standards for clinical, laboratory and genetic data, such as those standards integrated into the Unified Medical Language System Metathesaurus.

9. A method according to claim 1 wherein the first statistical model is trained on a second data set where the number of elements in the second set is large in comparison to the number of subjects in the second group used to train the model, and the model is trained on the sparse second data set using one or more shrinkage functions.

10. A method according to claim 1 wherein the first statistical model creates independent variables based on logical or arithmetic combinations of the first standardized data classes, and the first model is trained using one or more shrinkage functions.

11. A method according to claim 1 wherein the first statistical model creates independent variables based on logical or arithmetic combinations of the first standardized data classes, and the first model is trained using a log shrinkage function designed to minimize the amount of information in a regression parameters set.

12. A method according to claim 1 wherein the first statistical model is trained using the maximum likelihood method of logistic regression, together with a shrinkage function, where the shrinkage function is a log shrinkage function or an absolute magnitude shrinkage function.

13. A method according to claim 1 wherein the first statistical model for predicting the first phenotypic outcome, and the method of training the first statistical model, is inhaled from experts, such as clinical trial investigators, who publish their statistical models and the methods for training these statistical models according to a standardized electronic template so that the models and methods can be automatically inhaled and applied to the data.

14. A method according to claim 1 in which the first set of data includes genetic data for a sperm and egg cell, and the predicted outcome, or outcomes, is the probability of the progeny resulting from this sperm and egg cell having a particular phenotypic trait, or multiple traits.

15. A method according to claim 1 wherein the first set of data includes genetic data of parent organisms, or a parent organism, and the outcome is the probability of the progeny of these parent organisms having a particular phenotypic trait, or multiple phenotypic traits.

16. A method according to claim 1 wherein the first set of data includes genetic data for a set of three gametes resulting from a meiosis division, together with genetic data of the parent organism, from which is determined by the process of elimination the genetic information in a fourth preserved gamete;

and the first phenotypic outcome is the probability of the progeny derived from this preserved gamete having certain phenotypic features, for the purpose of determining whether the fourth preserved gamete is suitable for reproduction.

17. A method according to claim 1 wherein the first statistical model is continually retrained and refined based on newly inhaled and validated data contributing to the second data set.

18. A method according to claim 1 wherein the first data set invol es genetic mutation data, and the first statistical model invol es a decision tree.

19. A method according to claim 1 wherein the first data invol es genetic mutation data from the HIV viral genome, and the first statistical model invol es a decision tree.

20. A method for validating a first datum of genetic, phenotypic or clinical information for a first subject, based on a first set of genetic, phenotypic or clinical data from the first subject together with a second set of genetic, phenotypic or clinical data from a group of second subjects for whom the first datum of information is already known, the method comprising:

structuring the second set of data from the group of second subjects according to a first set of data classes that have unambiguous definition, and encompass the relevant genetic, phenotypic or clinical data, termed the first set of standardized data classes;

creating based on the second set of genetic, phenotypic or clinical data a computer executable first rule that determines the likelihood of the first datum, given the first set of standardized data classes;

structuring the first set of genetic, phenotypic or clinical data from the first individual according to a first set of data standardized data classes;

applying the first rule to estimate the likelihood of the first datum, based on the first set of data;

flagging the first set of data if the rule determines that the data is unlikely.

21. A method according to claim 20 wherein the computer-executable first rule is a statistical model that is trained using the second set of genetic, phenotypic or clinical data together with the known datum for each subject in the group of second subjects.

22. A method according to claim 20 wherein the first rule is a statistical model that is trained on sparse data using one or more shrinkage functions.

23. A method according to claim 20 wherein the first rule is a statistical model employing a decision tree.

24. A method according to claim 20 wherein the fist rule is a statistical model that is continually re-trained and updated using the most up-to-date second set of data from the most up-to-date second group of subjects.

25. A method according to claim 20 wherein the first or second set of genetic and phenotypic data is derived from multiple different external database systems, and is inhaled and formatted according to the set of standardized data classes by means of a cartridge designed to parse the data for each external system.

26. A method according to claim 20 wherein the standardized data classes are derived from a set of industry-wide standards for clinical, laboratory and genetic data, such as those standards integrated into the Unified Medical Language System Metathesaurus.

27. A method according to claim 20 wherein biometric authentication is used together with data level and fanctional level data access privileges in order for a patient, or a designated caregiver, to access or update data from the first or second set.

28. A method according to claim 20 wherein one or more elements of the first or second data are associated with a validator, which is the entity that validatated the correctness of that information, wherein the record indicating the reliability of the validator is maintained, and made available to entities who have to make clinical or market decisions based on the correctness of the validated data.

29. A method according to claim 20 wherein a potential buyer of the validated data is enabled to estimate the value of the validated information, based on the value of previous similar transactions, the quality of the data, and the record of the validator of that data.

30. A method according to claim 20 that enables the compensation of the entities who validated data in the first or second set, in the event that such data is sold based on a belief in the correctness of the data.

31. A method according to claim 20 wherein the first data is continually re-examined with the latest available computer-executable first rule which may relate to standards for data collection, or best clinical practice, and wherein electronic notification of data managers is enabled in order to take any actions necessary to keep the data compliant with the upto-date first rule.

32. A method according to claim 20 in which the first data relates to genetic sequence data from laboratories, and the rule validates that the genetic sample was not contaminated by another genetic sample by checking that the correlation of the aligned genetic sequence with other genetic sequences from the same laboratory does not exceed some threshold.

33. A method according to claim 20 in which the first data relates to genetic sequence data from laboratories, and the rule validates that the genetic sample was from the reported subject by checking the correlation of the aligned genetic sequence data with previous genetic sequence data from the same subject to ensure, within some probability threshold, that the sequences are sufficiently correlated to have arisen from the same subject.