Ancestry Determination Using Nested Hidden Markov Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current SNP genotyping methods fail to accurately identify which allele corresponds to which haplotype, leading to unordered pairs of alleles and difficulties in determining ethnic origin from genetic data.
Innovation Solution
The system employs hidden Markov models, specifically within-window diploid and inter-window HMMs, to assign labels to input genotype datasets by computing diploid HMMs for each chromosome window and building an inter-window HMM, using transition and emission probabilities to determine the likelihood of haplotype pairs and account for phasing errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If SNP genotyping is used to identify alleles, then genotype information can be obtained, but the system cannot determine which allele corresponds to which haplotype
Solution Approach 1:
The patent introduces population reference panels and hidden Markov models as intermediaries to infer haplotype phase information. These intermediaries bridge the gap between observed genotypes and unobserved haplotype configurations by leveraging patterns from reference populations and probabilistic modeling of genetic inheritance
Solution Approach 2:
The patent replaces direct mechanical observation of haplotype phase with computational inference methods. Instead of physically determining allele-haplotype correspondence through experimental means, the system uses statistical algorithms including hidden Markov models, Gibbs sampling, and Viterbi algorithms to compute the most likely phase assignments
2Measurement precision
If conventional methods like RFMix are used for ancestry determination, then ethnic origin can be estimated, but accuracy is insufficient
Solution Approach 1:
The patent implements a nested modeling structure where hidden Markov models are embedded within a broader ancestry inference framework. The HMMs model local haplotype patterns, which are then integrated into global ancestry composition estimates, creating multiple levels of nested probabilistic models that progressively refine ancestry determination
Solution Approach 2:
The patent divides the genome into segments or windows for independent analysis using hidden Markov models. Each segment is analyzed to determine local ancestry composition, and these segment-level results are then aggregated to produce overall ancestry estimates, enabling parallel processing and improved computational efficiency
3Measurement precision
If diploid HMMs are computed for each window, then labeling accuracy improves, but computational time increases
Solution Approach 1:
The patent segments the genome into windows and computes diploid hidden Markov models independently for each window. This segmentation enables parallel computation across multiple windows, reducing overall computational time while maintaining high labeling accuracy through localized probabilistic modeling of haplotype configurations
Solution Approach 2:
The patent performs preliminary computations of diploid HMMs for each window before integrating results into the final ancestry determination. By pre-computing window-specific probability distributions and storing them for later use, the system avoids redundant calculations during the final ancestry inference stage
Data Source
AI summary
An input genotype is divided into a plurality of windows, each including a sequence of SNPs. For each window, a diploid HMM is computed based on genotypes and/or phased haplotypes to determine a probability of a haplotype sequence being associated with a particular label. For example, the diploid HMM for a window is used to determine the emission probability that the window corresponds to a set of labels. An inter-window HMM, with a set of states for each window, is computed. Labels are assigned to the input genotype based on the inter-window HMM. Upper and lower bounds are estimated to produce a range of likely percentage values an input can be assigned to a given label. Confidence values are determined indicating a likelihood that an individual inherits DNA from a certain population. Maps are generated with polygons representing regions where a measure of ethnicity of population falls within specific ranges.