Scalable pipeline for local ancestry inference

1. An ancestry deconvolution system comprising:

one or more processors configured to:

obtain unphased genotype data of an individual;

phase the unphased genotype data to generate phased haplotype data;

use a learning machine to classify portions of the phased haplotype data as corresponding to specific ancestries respectively and generate initial classification results, wherein the learning machine includes a support vector machine (SVM); and

correct errors in the initial classification results to generate modified classification results; and

one or more memories coupled with the one or more processors, configured to provide the one or more processors with instructions.

2. The system of claim 1 , wherein the SVM uses a modified kernel.

3. The system of claim 1 , wherein the learning machine is trained using supervised learning.

4. The system of claim 1 , wherein to phase the unphased genotype data to generate phased haplotype data includes to perform out-of-sample phasing where the individual's genotype data is not included in a reference population.

5. An ancestry deconvolution system comprising:

one or more processors configured to:

obtain unphased genotype data of an individual;

phase the unphased genotype data to generate phased haplotype data;

use a learning machine to classify portions of the phased haplotype data as corresponding to specific ancestries respectively and generate initial classification results;

correct errors in the initial classification results to generate modified classification results; and

recalibrate the modified classification results; and

one or more memories coupled with the one or more processors, configured to provide the one or more processors with instructions.

6. The system of claim 5 , wherein the modified classification results include posterior probabilities, and to recalibrate the modified classification results includes to determine an ideal accuracy—posterior probabilities correspondence and an actual accuracy—posterior probabilities correspondence.

7. The system of claim 5 , wherein the modified classification results include posterior probabilities, and to recalibrate the modified classification results includes to recalibrate posterior probabilities using Platt's recalibration.

8. The system of claim 5 , wherein to recalibrate the modified classification results further includes to perform K-class recalibration.

9. The system of claim 5 , wherein the modified classification results include posterior probabilities, and to recalibrate the modified classification results includes to recalibrate posterior probabilities using isotonic regression.

10. The system of claim 1 , wherein the one or more processors are further configured to perform label clustering.

11. The system of claim 10 , wherein:

the modified classification results include posterior probabilities; and

to perform label clustering includes to:

determine whether a posterior probability with highest value meets a threshold; and

in the event that the threshold is not met, cluster at least some of the posterior probabilities to form a new probability associated with a broader geographical region.

12. A method of ancestry deconvolution, comprising:

obtaining unphased genotype data of an individual;

phasing, using one or more processors, the unphased genotype data to generate phased haplotype data;

using a learning machine to classify portions of the phased haplotype data as corresponding to specific ancestries respectively and generate initial classification results, wherein the learning machine includes a support vector machine (SVM); and

correcting errors in the initial classification results to generate modified classification results.

13. The method of claim 12 , wherein the SVM uses a modified kernel.

14. The method of claim 12 , wherein the learning machine is trained using supervised learning.

15. The method of claim 12 , wherein phasing the unphased genotype data to generate phased haplotype data includes performing out-of-sample phasing where the individual's genotype data is not included in a reference population.

16. A method of ancestry deconvolution, comprising:

obtaining unphased genotype data of an individual;

phasing, using one or more processors, the unphased genotype data to generate phased haplotype data;

using a learning machine to classify portions of the phased haplotype data as corresponding to specific ancestries respectively and generate initial classification results;

correcting errors in the initial classification results to generate modified classification results; and

recalibrating the modified classification results.

17. The method of claim 16 , wherein the modified classification results include posterior probabilities, and recalibrating the modified classification results includes determining an ideal accuracy—posterior probabilities correspondence and an actual accuracy—posterior probabilities correspondence.

18. The method of claim 16 , wherein the modified classification results include posterior probabilities, and recalibrating the modified classification results includes recalibrating posterior probabilities using Platt's recalibration.

19. The method of claim 16 , wherein recalibrating the modified classification results further includes performing K-class recalibration.

20. The method of claim 16 , wherein the modified classification results include posterior probabilities, and recalibrating the modified classification results includes recalibrating posterior probabilities using isotonic regression.

21. The method of claim 12 , further comprising performing label clustering.

22. The method of claim 21 , wherein:

the modified classification results include posterior probabilities; and

performing label clustering includes:

determining whether a posterior probability with highest value meets a threshold; and

in the event that the threshold is not met, clustering at least some of the posterior probabilities to form a new probability associated with a broader geographical region.

23. A computer program product for ancestry deconvolution, the computer program product being embodied in a tangible computer readable storage medium and comprising computer instructions for:

obtaining unphased genotype data of an individual;

phasing the unphased genotype data to generate phased haplotype data;

using a learning machine to classify portions of the phased haplotype data as corresponding to specific ancestries respectively and generate initial classification results, wherein the learning machine includes a support vector machine (SVM); and

correcting errors in the initial classification results to generate modified classification results.