Multidimensional heritability analysis of neuroanatomical shape
Jingwei Li
Brain Imaging Genetics
Genetic Variation
Behavior Cognition Neuroanatomy
Brain Imaging Genetics
Genetic Variation
Neuroanatomy
Descriptors of Brain Structures
โข One-dimensional descriptors (Hibar2015; Stein2012; Sabuncu2012)
โ Volume
โ Surface area
โ โฆ
โข Drawbacks
โ Limited when capturing the anatomical variation
Same area
Descriptors of Brain Structures
โข Multi-dimensional shape descriptor: truncated Laplace-Beltrami Spectrum (LBS)โข ๐: R๐ โ R๐+๐ is the local parametrization of a submonifold ๐ of R๐+๐
๐๐๐ =< ๐๐๐, ๐๐๐ >, ๐บ = ๐๐๐ ๐ร๐, ๐ = det๐บ, ๐๐๐ = ๐บโ1 ๐, ๐
โข If ๐ and ๐ are real-valued functions defined on ๐, then
๐ป ๐, ๐ = ๐,๐ ๐๐,๐ ๐๐๐ ๐๐๐, ฮ๐ =
1
๐ ๐,๐ ๐๐ ๐
๐๐๐ ๐๐๐
where ๐ป ๐, ๐ โ< grad ๐, grad ๐ > and ฮ๐ โ div grad ๐ .
โข Solve Laplacian eigenvalue problem: ฮ๐ = ๐๐Nabla operator Laplace-Beltrami operator
eigenfunction eigenvalue
Descriptors of Brain Structures
โข Multi-dimensional shape descriptor: truncated Laplace-Beltrami Spectrum (LBS)
Translate Laplacian eigenvalue problem: ๐ซ๐ = ๐๐ to a variational problem:
โข ๐ฮ๐ ๐๐ = โ ๐ป ๐, ๐ ๐๐
โข Since ๐ป ๐, ๐ = ๐,๐ ๐๐,๐ ๐๐๐ ๐๐๐ and ๐ฮ๐ ๐๐ = ๐ ๐๐ ๐๐ = โ๐ ๐๐๐๐
๐,๐ ๐๐,๐ ๐๐๐ ๐๐๐ ๐๐ = ๐ ๐๐๐๐
variational problem
Green formula
Descriptors of Brain Structures
โข Multi-dimensional shape descriptor: truncated Laplace-Beltrami Spectrum (LBS)
Discretization of ๐,๐ ๐๐,๐ ๐๐๐ ๐๐๐ ๐๐ = ๐ ๐๐๐๐:
โข Choose ๐ linearly independent form functions: ๐1 ๐ฅ , ๐2 ๐ฅ ,โฆ , ๐๐ ๐ฅ as basis functions (e.g. ๐ฅ, ๐ฅ2, ๐ฅ3, โฆ) defined on the parameter space.
โข Any eigenfunction ๐ can be approximately projected to the basis functions:๐ ๐ฅ โ ๐น ๐ฅ = ๐1๐1 ๐ฅ + โฏ+ ๐๐๐๐ ๐ฅ
โข To solve ๐ โ , substitute ๐ and ๐ โ into the variational problem.
โข Define ๐ด = ๐๐๐ ๐ร๐ = ๐,๐ ๐๐๐น๐ ๐๐๐น๐ ๐๐๐๐๐๐ร๐
and ๐ต =
๐๐๐ ๐ร๐ = ๐น๐๐น๐๐๐ ๐ร๐
=> ๐ด๐ = ๐๐ต๐General eigenvalue problem
Descriptors of Brain Structures
โข Multi-dimensional shape descriptor: truncated Laplace-Beltrami Spectrum (LBS)
โ Solve a Laplacian eigenvalue problem defined based on the brain region
โ Obtain the first ๐ eigenvalues
โข Properties (Reuter 2006):
โ Isometric invariant
โข For planar shapes and 3D-solids:isometry congruency(identical after rigid body transformation)
โข For surface:isometry โ congruency
Descriptors of Brain Structures
โข Multi-dimensional shape descriptor: truncated Laplace-Beltrami Spectrum (LBS)
โ Solve a Laplacian eigenvalue problem defined based on the brain region
โ Obtain the first ๐ eigenvalues
โข Properties (Reuter 2006):
โ Isometric invariant
โ scaling a n-dimensional manifold by the factor ๐ results in
scaled eigenvalues by the factor 1
๐2
โ โฆ In this paper, eigenvalues are scaled:
๐๐,๐ = ๐๐,๐ โ ๐๐2/3
๐: subject; ๐: dimension
Heritability
โข A phenotype/trait can be influenced by genetic and environmental effects.
โข Heritability: how much of the variation in a phenotype/trait is due to variation in genetic factors.
Main Idea of This Paper
โข Truncated LBS is more representative for a shape compared to volume.
โข Use truncated LBS as descriptors for 12 brain regions to compute heritability. Compare that with volume-based heritability.
โข To adapt truncated LBS into GCTA (Genome-wide Complex Trait Analysis) (Yang 2011) heritability model, propose a multi-dimensional heritability model.
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
Additive genetic component Common environmental component
Unique environmental component
๐ ร 1 trait vector(๐: #subjects)
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
K: genetic similarity matrix
โขFamilial study: ๐พ = 2 ร ๐พ๐๐๐ โ๐๐ ๐ถ๐๐๐๐๐๐๐๐๐๐ก๐ . E.g. parent-offspring (0.5), identical twins (1), full siblings (0.5), half siblings (0.25)
โขUnrelated subjects study: genome-side single-nucleotide polymorphism (SNP) data
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
What is Single-Nucleotide Polymorphism (SNP):
โข Each locus on a DNA sequence is a single nucleotide adenine (A), thymine (T), cytosine (C), or guanine (G).
โข SNP: a DNA sequence variation occurring when the types of single nucleotide in the genome (or other shared sequence) differs between individuals or paired chromosomes in one subject. E.g., AAGCCTA and AAGCTTA.
โข SNP can leads to alleles (variants of a given gene).
โข Each SNP can have 3 genotypes: AA, Aa, aa (denoted as 0-2)
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
How to compute genetic similarity from SNP:
โข ๐(#subjects x #SNPs).
โข Standardize each column of ๐ (mean 0, variance 1).
โข ๐พ =๐๐๐
#๐๐๐๐
0 โฏ 22 โฏ 1โฎ1 โฏ
โฎ0
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
ฮ: shared environment matrix between the subjects
โขFamilial study: e.g., twins & non-twin siblings (1)
โขUnrelated subjects study: ฮ vanishes
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
Identical matrix
GCTA heritability model
๐ฆ = ๐ + ๐ + ๐
๐~๐ 0, ๐๐ด2๐พ ๐~๐ 0, ๐๐ถ
2ฮ ๐~๐ 0, ๐๐ธ2๐ผ
โ2 =๐๐ด2
๐๐ด2 + ๐๐ถ
2 + ๐๐ธ2
โ2: the variance in the trait explained by the variance in additive genetic component
heritability
Multi-dimensional traits heritability model
๐ = ๐บ + ๐ถ + ๐ธ
๐ฃ๐๐ ๐บ ~๐ 0, ฮฃ๐ด โจ๐พ , ๐ฃ๐๐ ๐ถ ~๐ 0, ฮฃ๐ถ โจฮ , ๐ฃ๐๐ ๐ธ ~๐ 0, ฮฃ๐ธ โจ๐ผ
๐ ร๐ trait matrix(๐: #subjects)(๐: #dimensions)
ฮฃ๐ด = ๐๐ด๐๐ ๐ร๐: ๐๐ด๐๐ is
the genetic covariance between ๐-th and ๐ -thdimensions in traits
ฮฃ๐ถ = ๐๐ถ๐๐ ๐ร๐: ๐๐ถ๐๐ is
the common environmental covariance between ๐-thand ๐ -th dimensions in traits
ฮฃ๐ธ = ๐๐ธ๐๐ ๐ร๐: ๐๐ธ๐๐ is
the unique environmental covariance between ๐-thand ๐ -th dimensions in traits
Multi-dimensional traits heritability model
๐ = ๐บ + ๐ถ + ๐ธ
๐ฃ๐๐ ๐บ ~๐ 0, ฮฃ๐ด โจ๐พ , ๐ฃ๐๐ ๐ถ ~๐ 0, ฮฃ๐ถ โจฮ , ๐ฃ๐๐ ๐ธ ~๐ 0, ฮฃ๐ธ โจ๐ผ
โจ: Kronecker product
ฮฃ๐ด๐๐ โจ๐พ =
๐๐ด11๐พ ๐๐ด12๐พ โฏ ๐๐ด1๐๐พ
๐๐ด21๐พ ๐๐ด22๐พ โฏ ๐๐ด2๐๐พ
โฎ โฎ โฎ๐๐ด๐1๐พ ๐๐ด๐2๐พ โฏ ๐๐ด๐๐๐พ
Multi-dimensional traits heritability model
๐ = ๐บ + ๐ถ + ๐ธ
๐ฃ๐๐ ๐บ ~๐ 0, ฮฃ๐ด โจ๐พ , ๐ฃ๐๐ ๐ถ ~๐ 0, ฮฃ๐ถ โจฮ , ๐ฃ๐๐ ๐ธ ~๐ 0, ฮฃ๐ธ โจ๐ผ
๐ฃ๐๐ ๐1, ๐2, โฏ , ๐๐ =
๐1 ๐2โฎ ๐๐
Multi-dimensional traits heritability model
๐ = ๐บ + ๐ถ + ๐ธ
๐ฃ๐๐ ๐บ ~๐ 0, ฮฃ๐ด โจ๐พ , ๐ฃ๐๐ ๐ถ ~๐ 0, ฮฃ๐ถ โจฮ , ๐ฃ๐๐ ๐ธ ~๐ 0, ฮฃ๐ธ โจ๐ผ
โ2 =tr ฮฃ๐ด
tr ฮฃ๐ด + tr ฮฃ๐ถ + tr ฮฃ๐ธ=
๐=1
๐
๐พ๐โ๐2
where ๐พ๐ =๐๐ด๐๐ + ๐๐ถ๐๐ + ๐๐ธ๐๐
๐=1๐ ๐๐ด๐๐ + ๐๐ถ๐๐ + ๐๐ธ๐๐
โ๐2 =
๐๐ด๐๐
๐๐ด๐๐ + ๐๐ถ๐๐ + ๐๐ธ๐๐
The multi-dimensional trait heritability is a weighted average of the heritability of each dimension.
heritability
Multi-dimensional traits heritability model
โข Properties
โ Invariant to rotations of data
๐๐ = ๐บ๐ + ๐ถ๐ + ๐ธ๐
โ๐2 = โ2
๐ = ๐บ + ๐ถ + ๐ธ (1)
(2)
heritability from model (1)heritability from model (2)
๐๐๐ = ๐๐๐ = ๐ผ
Consider covariates
โข Sometimes, we want to study the effects after controlling some nuisance variables by regressing them out.
โข E.g., age, gender, handness
Consider covariates
๐ = ๐๐ต + ๐บ + ๐ถ + ๐ธ
๐ฃ๐๐ ๐บ ~๐ 0, ฮฃ๐ด โจ๐พ , ๐ฃ๐๐ ๐ถ ~๐ 0, ฮฃ๐ถ โจฮ , ๐ฃ๐๐ ๐ธ ~๐ 0, ฮฃ๐ธ โจ๐ผ
๐ = ๐๐๐ = ๐๐๐บ + ๐๐๐ถ + ๐๐๐ธ = ๐บ + ๐ถ + ๐ธ
๐ฃ๐๐ ๐บ ~๐ 0, ฮฃ๐ด โจ ๐๐๐พ๐ , ๐ฃ๐๐ ๐ถ ~๐ 0, ฮฃ๐ถ โจ ๐๐ฮ๐ ,
๐ฃ๐๐ ๐ธ ~๐ 0, ฮฃ๐ธ โจ๐ผ
๐๐๐ = 0๐๐๐ = ๐ผ
๐๐๐ = ๐ผ โ ๐ ๐๐๐ โ1๐๐
๐: ๐ ร ๐ โ ๐
Covariates (๐ ร ๐)
Analysis
โข Datasets:
โ Genomics Superstruct Project (GSP; N = 1320) โ unrelated subjects
โ Human Connectome Project (HCP; N = 590)โข 72 monozygotic twin pairs
โข 69 dizygotic twin pairs
โข 253 full siblings of twins
โข 55 singletons
โข 12 brain structures
โข Traits
โ Volume
โ Truncated LBS
Volume heritability (GSP data)
โข Before multiple comparisons correction: 3/12 brain structures are significantโข After multiple comparisons correction: none is significantโข Most structures: parametric & nonparametric p values are similar => standard errors
estimates are accurate
Volume heritability (GSP data)
Test-retest reliability:โข Linโs concordance correlation coefficient
๐๐ =2๐๐๐ฅ๐๐ฆ
๐๐ฅ2 + ๐๐ฆ
2 + ๐๐ฅ โ ๐๐ฆ2
variance mean
correlation coefficient
๐ฅ, ๐ฆ: use repeated runs on separate days of the same set of subjects
Truncated LBS heritability (GSP data)
โข Before multiple comparisons correction: 7/12 brain structures are significantโข After multiple comparisons correction: 5/12 brain structures are significantโข Most structures: parametric & nonparametric p values are similar => standard errors
estimates are accurateโข Smaller standard error than volume-based heritability
Truncated LBS heritability (GSP data)
Test-retest reliability:โข Averaged Linโs concordance correlation coefficient
across ๐ dimensions
๐๐ =2๐๐๐ฅ๐๐ฆ
๐๐ฅ2 + ๐๐ฆ
2 + ๐๐ฅ โ ๐๐ฆ2
variance mean
correlation coefficient
๐ฅ, ๐ฆ: use repeated runs on separate days of the same set of subjects
Truncated LBS heritability (GSP data)
Truncated LBS heritability (HCP data)
โข Only significant brain structures results are shown
โข Consistently higher than GSP datasetโ Possible reason: in unrelated subjects only the variation of some
common SNPs are captured.
Structure ๐๐ Standard Error
Accumbens area 0.309 0.162
Caudate 0.583 0.124
Cerebellum 0.653 0.120
Corpus Callosum 0.558 0.136
Hippocampus 0.363 0.190
Third Ventricle 0.536 0.134
Putamen 0.483 0.212
Visualizing principal mode of shape variation
โข PCA is a kind of rotation of data. The first PC of LBS explains a large percentage of shape variation.
โข Heritability model: (1) invariant to rotation; (2) heritability of multi-dimensional trait = weighted average of each dimensionโs heritability
โข The heritability of truncated LBS is the weighted average of the first M PCsโ heritability.
Visualizing principal mode of shape variationProcedures (for one brain structure)1. Register each subjectโs mask (1 โ in structure, 0 โ out of structure)
to a common used template.
2. Create a population average of structure surface for plottingโ A weighted average of all subjectsโ registered mask image
โ Weight: Gaussian kernel โข center: average of first PC
โข distance: subject-specific corresponding first PC <-> center
โข Width: resulting 500 shapes have non-0 weights
โ The isosurface with 0.5 in the averaged map
3. Use the same Gaussian kernel, generate averaged maps by including the shapes around +2 standard deviation of the first PC (-2 s.d. as well)
4. Plot the difference between the two maps in step 3 on the surface generated in step 2.
Visualizing principal mode of shape variation
Red: shapes around +2 s.d. are larger than -2 s.d.
Blue: shapes around -2 s.d. are larger than +2 s.d.
Strengths
โข Use truncated LBS instead of volume as features
โ Capture more shape variation
โ Isometry invariance
โ Does not require any registration or mapping (Reuter 2006 & 2009)
โข Generalize the concept of heritability into multi-dimensional phenotypes
โ Other applications (multi-tests of one behavior; disease study)
Strengths
โข Variability of heritability estimation
โ Multi-dimensional trait heritability model < original GCTA model (unrelated subject dataset)
โ Heritability estimates are more accurate, more significant
โข Propose a visualization method for shape variation
โ Interpretation: shape variation along the first PC axis of the shape descriptor
Weakness
โข Optimal number of eigenvalue may not be 50
โ Only 30, 50, 70 are tested
โ Error bars for difference number of eigenvalues are not shown
โ Other number except 50 (used in paper) could lead to higher heritability and smaller error bars
Weakness
โข Optimal number of eigenvalue can be different for different brain structures
โ Amygdala: heritability is similar for 30, 50, 70 eigenvalues (even decrease)
โ 3rd-ventricle: heritability increases from 0.4 to 0.6
Weakness
โข Links between proposed visualization method and LBS heritability are not clear.
โข Only volume-based GCTA heritability is compared to the new method and new model.
โ More comparisons with the literature (e.g., Gilmore 2010; Baare 2001)
Backup: invariant to rotations of data
๐๐๐ฃ ๐ฃ๐๐ ๐บ๐
= ๐๐๐ฃ ๐๐โจ๐ผ ๐ฃ๐๐ ๐บ
= ๐๐โจ๐ผ ๐ฃ๐๐ ๐บ ๐โจ๐ผ
= ๐๐โจ๐ผ ฮฃ๐ดโจ๐พ ๐โจ๐ผ
= ๐๐ฮฃ๐ด๐ โจ๐พ
Similarly, ๐๐๐ฃ ๐ฃ๐๐ ๐ถ๐ = ๐๐ฮฃ๐ถ๐ โจฮ, ๐๐๐ฃ ๐ฃ๐๐ ๐ธ๐ = ๐๐ฮฃ๐ธ๐ โจ๐ผ
โ๐2 =
๐ก๐ ๐๐ฮฃ๐ด๐
๐ก๐ ๐๐ฮฃ๐ด๐ + ๐ก๐ ๐๐ฮฃ๐ถ๐ + ๐ก๐ ๐๐ฮฃ๐ธ๐
=๐ก๐ ฮฃ๐ด ๐๐๐
๐ก๐ ฮฃ๐ด ๐๐๐ + ๐ก๐ ฮฃ๐ถ ๐๐๐ + ๐ก๐ ฮฃ๐ธ(๐๐๐)
=๐ก๐ ฮฃ๐ด
๐ก๐ ฮฃ๐ด + ๐ก๐ ฮฃ๐ถ + ๐ก๐ ฮฃ๐ธ= โ2
Theorem: ๐ฃ๐๐ ๐ด๐๐ต = ๐ต๐โจ๐ด ๐ฃ๐๐ ๐Here ๐ด = ๐ผ, ๐ = ๐บ, ๐ต = ๐
โข ๐ดโจ๐ต ๐ = ๐ด๐โจ๐ต๐
โข ๐๐๐ฃ ๐ด๐ = ๐ด๐๐๐ฃ ๐ ๐ด๐
๐ดโจ๐ต ๐ถโจ๐ท = ๐ด๐ถโจ๐ต๐ท
โข ๐ก๐ ๐ด๐ต๐ถ = ๐ก๐ ๐ต๐ถ๐ด = ๐ก๐ ๐ถ๐ด๐ตโข Associative property of matrix
multiplication
Backup: multi-dimensional trait heritability is a weighted average of heritability of each dimension
โ2 =๐ก๐ ฮฃ๐ด
๐ก๐ ฮฃ๐ด + ฮฃ๐ถ + ฮฃ๐ธ
= ๐=1๐ ๐๐ด๐๐
๐=1๐ ๐๐ด๐๐ + ๐=1
๐ ๐๐ถ๐๐ + ๐=1๐ ๐๐ธ๐๐
=
๐=1
๐๐๐ด๐๐ + ๐๐ถ๐๐ + ๐๐ธ๐๐
๐=1๐ ๐๐ด๐๐ + ๐๐ถ๐๐ + ๐๐ธ๐๐
โ ๐๐ด๐๐
๐๐ด๐๐ + ๐๐ถ๐๐ + ๐๐ธ๐๐
=
๐=1
๐
๐พ๐โ๐2
Backup: moment-matching estimator for unrelated subjects (no shared environmental component)๐๐๐ฃ ๐ฆ๐ , ๐ฆ๐ = ๐๐ด๐๐ ๐พ + ๐๐ธ๐๐ ๐ผ โน ๐ฆ๐๐ฆ๐
๐ = ๐๐ด๐๐ ๐พ + ๐๐ธ๐๐ ๐ผ
To estimate ๐๐ด๐๐ , ๐๐ธ๐๐ , use a regression model:
๐ฃ๐๐ ๐ฆ๐๐ฆ๐ ๐ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ
โน ๐ฆ๐ โจ๐ฆ๐ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ
โน ๐ฃ๐๐ ๐พ ๐ ๐ฆ๐ โจ๐ฆ๐ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ ๐๐ฃ๐๐ ๐พ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐พ ๐๐ฃ๐๐ ๐ผ
๐ฃ๐๐ ๐ผ ๐ ๐ฆ๐ โจ๐ฆ๐ = ๐๐ด๐๐ ๐ฃ๐๐ ๐ผ ๐๐ฃ๐๐ ๐พ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ ๐๐ฃ๐๐ ๐ผ
โน ๐ฆ๐ โจ๐ฆ๐
๐๐ฃ๐๐ ๐พ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ ๐๐ฃ๐๐ ๐พ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ ๐๐ฃ๐๐ ๐พ
๐ฆ๐ โจ๐ฆ๐๐๐ฃ๐๐ ๐ผ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ ๐๐ฃ๐๐ ๐ผ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ ๐๐ฃ๐๐ ๐ผ
โน ๐ฆ๐ ๐โจ๐ฆ๐
๐ ๐ฃ๐๐ ๐พ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ ๐๐ฃ๐๐ ๐พ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ ๐๐ฃ๐๐ ๐พ
๐ฆ๐ ๐โจ๐ฆ๐
๐ ๐ฃ๐๐ ๐ผ = ๐๐ด๐๐ ๐ฃ๐๐ ๐พ ๐๐ฃ๐๐ ๐ผ + ๐๐ธ๐๐ ๐ฃ๐๐ ๐ผ ๐๐ฃ๐๐ ๐ผ
โน ๐ฆ๐๐๐พ๐ฆ๐ = ๐๐ด๐๐ ๐ก๐ ๐พ
2 + ๐๐ธ๐๐ ๐ก๐ ๐พ
๐ฆ๐๐๐ฆ๐ = ๐๐ด๐๐ ๐ก๐ ๐พ + ๐๐ธ๐๐ ๐ก๐ ๐ผ
Backup: moment-matching estimator for unrelated subjects (no shared environmental component)
โน๐๐ด๐๐ ๐๐ธ๐๐
=๐ก๐ ๐พ2 ๐ก๐ ๐พ
๐ก๐ ๐พ ๐ก๐ ๐ผ
โ1๐ฆ๐๐๐พ๐ฆ๐ ๐ฆ๐๐๐ฆ๐
โน
๐๐ด๐๐ =๐ฆ๐๐ ๐๐พ โ ๐ก๐ ๐พ ๐ผ ๐ฆ๐ ๐๐ก๐ ๐พ2 โ ๐ก๐2 ๐พ
โ๐ฆ๐๐ ๐พ โ ๐๐ผ ๐ฆ๐
๐๐พ
๐๐ธ๐๐ =๐ฆ๐๐ ๐ก๐ ๐พ2 ๐ผ โ ๐ก๐ ๐พ ๐พ ๐ฆ๐ ๐๐ก๐ ๐พ2 โ ๐ก๐2[๐พ]
=๐ฆ๐๐ ๐ ๐ผ โ ๐๐พ ๐ฆ๐
๐๐พ
where ๐ = ๐ก๐ ๐พ๐, ๐ = ๐ก๐ ๐พ2
๐, ๐๐พ = ๐ก๐ ๐พ2 โ ๐ก๐2 ๐พ๐ = ๐ ๐ โ ๐
โน ฮฃ๐ด =๐๐ ๐พ โ ๐๐ผ ๐
๐๐พ, ฮฃ๐ธ =
๐๐ ๐ ๐ผ โ ๐๐พ ๐
๐๐พ
Backup: sampling variance of the point estimator
๐๐ด โ๐พ โ ๐๐ผ
๐๐พ, ๐๐ธ โ
๐ ๐ผ โ ๐๐พ
๐๐พ
๐ก๐ด โ ๐ก๐ ฮฃ๐ด = ๐ก๐ ๐๐๐๐ด๐ , ๐ก๐ธ = ๐ก๐ ฮฃ๐ธ = ๐ก๐ ๐๐๐๐ธ๐ , ๐ก =๐ก๐ด๐ก๐ธ
The heritability is a function of ๐ก: ๐ ๐ก =๐ก๐ด
๐ก๐ด+๐ก๐ธ
๐ฃ๐๐ โ๐๐๐2 = ๐ฃ๐๐ ๐ ๐ก โ
๐๐ ๐ก
๐๐ก๐๐๐ฃ ๐ก
๐๐ ๐ก
๐๐ก๐
where ๐๐ ๐ก
๐๐ก=
๐๐ ๐ก
๐๐ก,๐๐ ๐ก
๐๐ก=
๐ก๐ธ
๐ก๐ด+๐ก๐ธ2 ,
โ๐ก๐ด
๐ก๐ด+๐ก๐ธ2
Define ๐๐๐ = ๐๐๐ฃ ๐ฆ๐ , ๐ฆ๐ = ๐๐ด๐๐ ๐พ + ๐๐ธ๐๐ ๐ผ
Backup: sampling variance of the point estimator
๐๐๐ฃ ๐ก๐ ๐๐๐๐ผ๐ , ๐ก๐ ๐๐๐๐ฝ๐
= ๐,๐ =1
๐
๐๐๐ฃ ๐ฆ๐๐๐๐ผ๐ฆ๐ , ๐ฆ๐
๐๐๐ฝ๐ฆ๐
= 2 ๐,๐ =1
๐
๐ก๐ ๐๐ผ๐๐๐ ๐๐ฝ๐๐๐
โน ๐๐๐ฃ ๐ก = 2 ๐,๐ =1
๐ ๐ก๐ ๐๐ด๐๐๐ ๐๐ด๐๐๐ ๐ก๐ ๐๐ด๐๐๐ ๐๐ธ๐๐๐ ๐ก๐ ๐๐ธ๐๐๐ ๐๐ด๐๐๐ ๐ก๐ ๐๐ธ๐๐๐ ๐๐ธ๐๐๐
โ 2 ๐,๐ =1
๐
๐๐ด๐๐ + ๐๐ธ๐๐ 2 ๐ก๐ ๐๐ด
2 ๐ก๐ ๐๐ด๐๐ธ๐ก๐ ๐๐ธ๐๐ด ๐ก๐ ๐๐ธ
2
=2๐ก๐ ฮฃ๐ด + ฮฃ๐ธ
2
๐๐พ
1 โ๐โ๐ ๐
โ2๐ก๐ ฮฃ๐ด + ฮฃ๐ธ
2
๐๐พ
1 โ1โ1 1
Quadratic form of statistics:๐๐๐ฃ ๐๐ฮ1๐, ๐
๐ฮ2๐ = 2๐ก๐ ฮ1ฮฃฮ2ฮฃ + 4๐๐ฮ1ฮฃฮ2๐Here ๐ = 0
๐๐๐ = ๐๐ด๐๐ ๐พ + ๐๐ธ๐๐ ๐ผ
โ ๐๐ด๐๐ ๐ผ + ๐๐ธ๐๐ ๐ผ
๐พ โ ๐ผโน ๐ โ 1, ๐ โ 1
Backup: sampling variance of the point estimator
๐ก๐ ๐๐ด2 = ๐ก๐
๐พ โ ๐๐ผ 2
๐๐พ= ๐ก๐
๐พ โ๐ก๐ ๐พ๐
๐ผ2
๐ก๐ ๐พ2 โ๐ก๐2 ๐พ๐
2 = ๐ก๐๐พ2 โ 2
๐ก๐ ๐พ๐ ๐พ๐ผ +
๐ก๐2 ๐พ๐2 ๐ผ
๐ก๐ ๐พ2 โ๐ก๐2 ๐พ๐
2
=๐ก๐ ๐พ2 โ 2
๐ก๐2 ๐พ๐ +
๐ก๐2 ๐พ๐
๐ก๐ ๐พ2 โ๐ก๐2 ๐พ๐
2 =1
๐๐พ
๐ก๐ ๐๐ด๐๐ธ = ๐ก๐๐พ โ ๐๐ผ ๐ ๐ผ โ ๐๐พ
๐๐พ2 =
๐ก๐ ๐ ๐พ๐ผ โ ๐๐พ2 โ ๐๐พ๐ผ2 + ๐2๐ผ๐พ
๐๐พ2
=
๐ก๐ ๐พ2
๐ ๐ก๐ ๐พ โ๐ก๐ ๐พ๐ ๐ก๐ ๐พ2 โ
๐ก๐2 ๐พ๐ +
๐ก๐3 ๐พ๐2
๐๐พ2
=
๐ก๐ ๐พ๐
๐ก๐ ๐พ2 โ ๐ก๐ ๐พ2 โ ๐ก๐ ๐พ +๐ก๐2 ๐พ๐
๐ก๐ ๐พ2 โ๐ก๐2 ๐พ๐
2 = โ๐
๐๐พ
Backup: sampling variance of the point estimator
๐ก๐ ๐๐ธ2 =
๐ก๐ ๐ ๐ผ โ ๐๐พ 2
๐๐พ2 =
๐ก๐ ๐ 2๐ผ โ 2๐ ๐๐พ + ๐2๐พ2
๐๐พ2
=๐
๐ก๐ ๐พ2
๐ ๐ โ 2๐ก๐2 ๐พ๐ +
๐ก๐2 ๐พ๐2
๐๐ก๐ ๐พ2 ๐ก๐ ๐พ2
๐๐พ2
=๐ ๐ก๐ ๐พ2 โ 2
๐ก๐2 ๐พ๐ +
๐ก๐2 ๐พ๐
๐๐พ ๐ก๐ ๐พ2 โ๐ก๐2 ๐พ๐
=๐
๐๐พ
๐ฃ๐๐ โ๐๐๐2 = ๐ฃ๐๐ ๐ ๐ก โ
๐๐ ๐ก
๐๐ก๐๐๐ฃ ๐ก
๐๐ ๐ก
๐๐ก๐
โ2๐ก๐ ฮฃ๐ด + ฮฃ๐ธ
2
๐๐พ ๐ก๐ด + ๐ก๐ธ 4๐ก๐ธ , โ๐ก๐ด
1 โ1โ1 1
๐ก๐ธโ๐ก๐ด
=2๐ก๐ ฮฃ๐ด + ฮฃ๐ธ
2
๐๐พ ๐ก๐ด + ๐ก๐ธ 4๐ก๐ด + ๐ก๐ธ
2
=2๐ก๐ ฮฃ๐ด + ฮฃ๐ธ
2
๐๐พ ๐ก๐ ฮฃ๐ด + ๐ก๐ ฮฃ๐ธ2=
2
๐๐พโ ๐ก๐ ฮฃ๐ด + ฮฃ๐ธ
2
๐ก๐ ฮฃ๐ด + ฮฃ๐ธ2=
2
๐๐พโ ๐ก๐ ฮฃ๐
2
๐ก๐ ฮฃ๐2
Backup: sampling variance of the point estimator
For univariate trait, ๐ก๐ ฮฃ๐2 = ๐ก๐2 ฮฃ๐ , โน ๐ฃ๐๐ โ๐๐๐
2 =2
๐๐พ
For multi-dimensional trait,
๐ก๐ ฮฃ๐2
๐ก๐2 ฮฃ๐=
๐=1๐ ๐๐
2
๐=1๐ ๐๐
2 โค 1 โน ๐ฃ๐๐ โ๐๐๐2 โค
2
๐๐พ