Bayesian variable selection regression. More...

#include <MuGen.h>

Inheritance diagram for BetaGrpBVSR:

Collaboration diagram for BetaGrpBVSR:

Public Member Functions
	BetaGrpBVSR ()
	Default constructor.

	BetaGrpBVSR (const Grp &y, const SigmaI &SigI, const string &predFlNam, const double &Nmul, const double &rSqMax, RanIndex &up, const string &outFlNam, const int &nThr)
	Basic constructor. More...

	BetaGrpBVSR (const Grp &y, const SigmaI &SigI, const string &predFlNam, const double &Nmul, const double &rSqMax, RanIndex &low, RanIndex &up, const string &outFlNam, const int &nThr)
	Basic constructor with replication. More...

	BetaGrpBVSR (const Grp &y, const SigmaI &SigI, const string &predFlNam, const double &Nmul, const double &rSqMax, const double &absLab, RanIndex &up, const string &outFlNam, const int &nThr)
	Basic constructor with missing data. More...

	BetaGrpBVSR (const Grp &y, const SigmaI &SigI, const string &predFlNam, const double &Nmul, const double &rSqMax, const double &absLab, RanIndex &low, RanIndex &up, const string &outFlNam, const int &nThr)
	Basic constructor with replication and missing data. More...

	~BetaGrpBVSR ()
	Destructor.

void	save (const SigmaI &SigI)
	Regression value store. More...

Public Member Functions inherited from BetaGrpFt
	BetaGrpFt ()
	Default constructor.

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const int &nThr)
	Simple constructor. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, RanIndex &up, const int &nThr)
	Simple constructor with a prior index. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, RanIndex &low, RanIndex &up, const int &nThr)
	Simple constructor with a prior index and replication. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const string &outFlNam, const int &nThr)
	Simple constructor with output file name. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, RanIndex &up, const string &outFlNam, const int &nThr)
	Simple constructor with a prior index and output file name. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, RanIndex &low, RanIndex &up, const string &outFlNam, const int &nThr)
	Simple constructor with a prior index, replication and output file name. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const double &absLab, const int &nThr)
	Missing data constructor. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const double &absLab, RanIndex &up, const int &nThr)
	Missing data constructor with a prior index. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const double &absLab, RanIndex &low, RanIndex &up, const int &nThr)
	Missing data constructor with a prior index and replication. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const double &absLab, const string &outFlNam, const int &nThr)
	Missing data constructor with output file name. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const double &absLab, RanIndex &up, const string &outFlNam, const int &nThr)
	Missing data constructor with a prior index and output file name. More...

	BetaGrpFt (const Grp &rsp, const string &predFlNam, const size_t &Npred, const double &absLab, RanIndex &low, RanIndex &up, const string &outFlNam, const int &nThr)
	Missing data constructor with a prior index, replication and output file name. More...

	BetaGrpFt (const Grp &rsp, const SigmaI &SigI, const string &predFlNam, const size_t &Npred, const double &Nmul, const double &rSqMax, RanIndex &up, const string &outFlNam, const int &nThr)
	Selection constructor. More...

	BetaGrpFt (const Grp &rsp, const SigmaI &SigI, const string &predFlNam, const size_t &Npred, const double &Nmul, const double &rSqMax, RanIndex &low, RanIndex &up, const string &outFlNam, const int &nThr)
	Selection constructor with replication. More...

	BetaGrpFt (const Grp &rsp, const SigmaI &SigI, const string &predFlNam, const size_t &Npred, const double &Nmul, const double &rSqMax, const double &absLab, RanIndex &up, const string &outFlNam, const int &nThr)
	Selection constructor with missing predictor data. More...

	BetaGrpFt (const Grp &rsp, const SigmaI &SigI, const string &predFlNam, const size_t &Npred, const double &Nmul, const double &rSqMax, const double &absLab, RanIndex &low, RanIndex &up, const string &outFlNam, const int &nThr)
	Selection constructor with missing predictor data and replication. More...

virtual	~BetaGrpFt ()
	Destructor.

	BetaGrpFt (const BetaGrpFt &mG)
	Copy constructor. More...

BetaGrpFt &	operator= (const BetaGrpFt &mG)
	Assignment operator. More...

virtual const gsl_matrix *	fMat () const
	Access to the fitted value matrix. More...

void	dump ()
	Dump to a file. More...

double	lnOddsRat (const Grp &y, const SigmaI &SigI, const size_t i) const
	Log-odds ratio. More...

void	update (const Grp &dat, const SigmaI &SigIm)
	Gaussian likelihood, improper prior. More...

void	update (const Grp &dat, const Qgrp &q, const SigmaI &SigIm)
	Student- $t$ likelihood, improper prior. More...

virtual void	update (const Grp &dat, const SigmaI &SigIm, const SigmaI &SigIp)
	Gaussian likelihood, 0-mean Gaussian prior. More...

virtual void	update (const Grp &dat, const Qgrp &q, const SigmaI &SigIm, const SigmaI &SigIp)
	Student- $t$ likelihood, 0-mean Gaussian prior. More...

virtual void	update (const Grp &dat, const SigmaI &SigIm, const Qgrp &qPr, const SigmaI &SigIp)
	Gaussian likelihood, 0-mean Student- $t$ prior. More...

virtual void	update (const Grp &dat, const Qgrp &q, const SigmaI &SigIm, const Qgrp &qPr, const SigmaI &SigIp)
	Student- $t$ likelihood, 0-mean Student- $t$ prior. More...

virtual void	update (const Grp &dat, const SigmaI &SigIm, const Grp &muPr, const SigmaI &SigIp)
	Gaussian likelihood, non-zero mean Gaussian prior. More...

virtual void	update (const Grp &dat, const Qgrp &q, const SigmaI &SigIm, const Grp &muPr, const SigmaI &SigIp)
	Student- $t$ likelihood, non-zero mean Gaussian prior. More...

virtual void	update (const Grp &dat, const SigmaI &SigIm, const Grp &muPr, const Qgrp &qPr, const SigmaI &SigIp)
	Gaussian likelihood, non-zero mean Student- $t$ prior. More...

virtual void	update (const Grp &dat, const Qgrp &q, const SigmaI &SigIm, const Grp &muPr, const Qgrp &qPr, const SigmaI &SigIp)
	Student- $t$ likelihood, non-zero mean Student- $t$ prior. More...

Public Member Functions inherited from Grp
virtual	~Grp ()
	Destructor.

virtual void	save ()
	Save to pre-specified file. More...

virtual void	save (const string &outFlNam)
	Save to file. More...

virtual void	save (const string &outMuFlNam, const string &outSigFlNam, const SigmaI &SigI)
	Joint save. More...

virtual void	save (const Grp &y, const SigmaI &SigI)
	Save with data and inverse-covariance. More...

void	mhlSave (const string &outFlNam, const SigmaI SigI)
	Save Mahalanobis distance. More...

const vector< MVnorm * > &	dataVec () const
	Get vector of row pointers. More...

virtual const gsl_matrix *	dMat () const
	Access the value matrix. More...

const size_t	Ndata () const
	Get number of rows. More...

const size_t	phenD () const
	Get number of traits. More...

const MVnorm *	operator[] (const size_t i) const
	Subscript operator. More...

MVnorm *	operator[] (const size_t i)
	Subscript operator. More...

virtual MuGrp	mean (RanIndex &grp)
	Group mean. More...

virtual const MuGrp	mean (RanIndex &grp) const
	Group mean. More...

virtual MuGrp	mean (RanIndex &grp, const Qgrp &q)
	Group weighted mean. More...

virtual const MuGrp	mean (RanIndex &grp, const Qgrp &q) const
	Group weighted mean. More...

void	center ()
	Center the value matrix. More...

Protected Member Functions
void	_updateFitted ()
	Update fitted values.

void	_ldToss (const gsl_vector var, const gsl_permutation prm, const double &rSqMax, const size_t &Nsel, const size_t &Npck, vector< vector< size_t > > &idx, vector< vector< size_t > > &rLd, gsl_matrix *Xpck)
	Testing candidates for correlation. More...

double	_MGkernel (const Grp &dat, const SigmaI &SigI) const
	Gaussian kernel. More...

double	_MGkernel (const Grp &dat, const SigmaI &SigI, const size_t &prInd) const
	Gaussian kernel dropping one predictor. More...

double	_MGkernel (const Grp &dat, const SigmaI &SigIe, const SigmaI &SigIpr, const size_t &prInd)
	Gaussian kernel adding one predictor. More...

Protected Member Functions inherited from BetaGrpFt
void	_rankPred (const gsl_matrix y, const SigmaI &SigI, gsl_vector XtX, gsl_permutation *prm)
	Rank predictors. More...

void	_rankPred (const gsl_matrix y, const SigmaI &SigI, const double &absLab, gsl_vector XtX, gsl_permutation *prm)
	Rank predictors with missing data. More...

void	_ldToss (const gsl_vector var, const gsl_permutation prm, const double &rSqMax, const size_t &Npck, vector< vector< size_t > > &idx, vector< vector< size_t > > &rLd, gsl_matrix *Xpck)
	Testing candidates for correlation. More...

Protected Member Functions inherited from Grp
	Grp ()

Protected Attributes
gsl_matrix *	_selB
	Matrix of selected predictors. More...

gsl_matrix *	_tmpXb
	$\boldsymbol{\mathrm{x}}_{\cdot l}\boldsymbol{\beta}_{l \cdot}$ matrix for the candidate dropped/added element

Protected Attributes inherited from BetaGrpFt
vector< vector< double > >	_fittedEach
	Partial fitted value matrices. More...

gsl_matrix *	_fittedAll
	Matrix of fitted values. More...

gsl_matrix *	_valueSum
	Sample storage matrix. More...

gsl_matrix *	_Xmat
	Predictor matrix. More...

int	_nThr
	Number of threads.

double	_numSaves
	Number of saves. More...

Protected Attributes inherited from Grp
vector< MVnorm * >	_theta
	Vector of pointers to value rows. More...

gsl_matrix *	_valueMat
	Value matrix. More...

RanIndex *	_lowLevel
	Lower level index. More...

RanIndex *	_upLevel
	Upper level index. More...

vector< gsl_rng * >	_rV
	Vector of PNG pointers. More...

string	_outFlNam
	Name of the output file.

Friends
void	RanIndexVS::update (const Grp &, const SigmaI &, BetaGrpBVSR *, const SigmaI &)

Detailed Description

Bayesian variable selection regression.

A multivariate implementation of Bayesian variable selection regression (BVSR), together with the RanIndexVS class. The model is similar to that described in [guan11] , but unlike their method a pre-selection of predictors based on a first-pass single-predictor regression is made. The user has control over the number of predictors to choose, and the cut-off for correlation among predictors (which arises as linkage disequilibrium in SNP regressions). This numebr is expressed as a multiple of genetic sample size. Once the pre-selected group of predictors is formed, variable selection is performed within the smaller group. Predictors discarded for correlation with picked variables are assigned the same probability of retention in the model as their corresponding picked predictors. Rao-Blackwellised values of regression coefficients are saved by the dump() function.

Warning: This class is experimental and has not been extensively tested

Constructor & Destructor Documentation

◆ BetaGrpBVSR() [1/4]

BetaGrpBVSR::BetaGrpBVSR	(	const Grp &	y,
		const SigmaI &	SigI,
		const string &	predFlNam,
		const double &	Nmul,
		const double &	rSqMax,
		RanIndex &	up,
		const string &	outFlNam,
		const int &	nThr
	)

Basic constructor.

Constructor with no missing data in predictors and no replication. Performs the initial pre-selection of predictors.

Parameters

[in]	Grp&	data
[in]	SigmaI&	data inverse-covariance
[in]	string&	preddictor file name
[in]	double&	number of predictors to keep, in multiples of sample size
[in]	double&	$r^2$ cut-off for predictor correlation
[in]	RanIndex&	prior index
[in]	string&	output file name
[in]	int&	number of threads

◆ BetaGrpBVSR() [2/4]

BetaGrpBVSR::BetaGrpBVSR	(	const Grp &	y,
		const SigmaI &	SigI,
		const string &	predFlNam,
		const double &	Nmul,
		const double &	rSqMax,
		RanIndex &	low,
		RanIndex &	up,
		const string &	outFlNam,
		const int &	nThr
	)

Basic constructor with replication.

Constructor with no missing data in predictors and replication. A RanIndex index is used instead of a design matrix. Performs the initial pre-selection of predictors.

Parameters

[in]	Grp&	data
[in]	SigmaI&	data inverse-covariance
[in]	string&	preddictor file name
[in]	double&	number of predictors to keep, in multiples of sample size
[in]	double&	$r^2$ cut-off for predictor correlation
[in]	RanIndex&	replicate (data) index
[in]	RanIndex&	prior index
[in]	string&	output file name
[in]	int&	number of threads

◆ BetaGrpBVSR() [3/4]

BetaGrpBVSR::BetaGrpBVSR	(	const Grp &	y,
		const SigmaI &	SigI,
		const string &	predFlNam,
		const double &	Nmul,
		const double &	rSqMax,
		const double &	absLab,
		RanIndex &	up,
		const string &	outFlNam,
		const int &	nThr
	)

Basic constructor with missing data.

Constructor with missing data in predictors and no replication. Missing data in the selected predictors are filled in using mean imputation. Performs the initial pre-selection of predictors.

Parameters

[in]	Grp&	data
[in]	SigmaI&	data inverse-covariance
[in]	string&	preddictor file name
[in]	double&	number of predictors to keep, in multiples of sample size
[in]	double&	$r^2$ cut-off for predictor correlation
[in]	double&	missing data label
[in]	RanIndex&	prior index
[in]	string&	output file name
[in]	int&	number of threads

◆ BetaGrpBVSR() [4/4]

BetaGrpBVSR::BetaGrpBVSR	(	const Grp &	y,
		const SigmaI &	SigI,
		const string &	predFlNam,
		const double &	Nmul,
		const double &	rSqMax,
		const double &	absLab,
		RanIndex &	low,
		RanIndex &	up,
		const string &	outFlNam,
		const int &	nThr
	)

Basic constructor with replication and missing data.

Constructor with missing data in predictors and replication. Missing data in the selected predictors are filled in using mean imputation. A RanIndex index is used instead of a design matrix. Performs the initial pre-selection of predictors.

Parameters

[in]	Grp&	data
[in]	SigmaI&	data inverse-covariance
[in]	string&	preddictor file name
[in]	double&	number of predictors to keep, in multiples of sample size
[in]	double&	$r^2$ cut-off for predictor correlation
[in]	double&	missing data label
[in]	RanIndex&	replicate (data) index
[in]	RanIndex&	prior index
[in]	string&	output file name
[in]	int&	number of threads

Member Function Documentation

◆ _ldToss()

void BetaGrpBVSR::_ldToss	(	const gsl_vector *	var,
		const gsl_permutation *	prm,
		const double &	rSqMax,
		const size_t &	Nsel,
		const size_t &	Npck,
		vector< vector< size_t > > &	idx,
		vector< vector< size_t > > &	rLd,
		gsl_matrix *	Xpck
	)

protected

Testing candidates for correlation.

Goes through the list of "top" predictors and eliminates lower-ranked candidates correlated with them. If any candidates are eliminated, the list of top predictors is augmented with predictors previously discarded. The identity of the predictors eliminated for correlation is saved. The correlation among predictors arises, for example, when estimating SNP effect in genetics (GWAS).

Warning: Tested only superficially

Parameters

[in]	gsl_vector*	$\left( \boldsymbol{x}^T\boldsymbol{x} \right)^{-1}$
[in]	gsl_permutation*	predictor ranks
[in]	double&	$r^2$ cut-off
[in]	size_t&	number of predictors to pick
[out]	vector<	vector<size_t> >& index of picked predictors
[out]	vector<	vector<size_t> >& index relating dropped correlated predictors to their group-defining predictor
[out]	gsl_matrix*	matrix of picked predictor values

◆ _MGkernel() [1/3]

double BetaGrpBVSR::_MGkernel	(	const Grp &	dat,
		const SigmaI &	SigI
	)		const

protectedvirtual

Gaussian kernel.

Calculates the multivariate Gaussian kernel value for all regression coefficients in the model.

Parameters

[in]	Grp&	data
[in]	SigmaI&	inverse-covariance

Returns: double kernel value

Reimplemented from BetaGrpFt.

◆ _MGkernel() [2/3]

double BetaGrpBVSR::_MGkernel	(	const Grp &	dat,
		const SigmaI &	SigI,
		const size_t &	prInd
	)		const

protectedvirtual

Gaussian kernel dropping one predictor.

Calculates the multivariate Gaussian kernel value for all regression coefficients in the model, except for the one indicated.

Parameters

[in]	Grp&	data
[in]	SigmaI&	inverse-covariance
[in]	size_t&	index of the dropped predictor

Returns: double kernel value

Reimplemented from BetaGrpFt.

◆ _MGkernel() [3/3]

double BetaGrpBVSR::_MGkernel	(	const Grp &	dat,
		const SigmaI &	SigIe,
		const SigmaI &	SigIpr,
		const size_t &	prInd
	)

protected

Gaussian kernel adding one predictor.

Calculates the multivariate Gaussian kernel value for all regression coefficients in the model, plus the indicated extra predictor with a Gaussian zero-mean prior.

Parameters

[in]	Grp&	data
[in]	SigmaI&	data inverse-covariance
[in]	SigmaI&	prior inverse-covariance
[in]	size_t&	index of the dropped predictor

Returns: double kernel value

◆ save()

void BetaGrpBVSR::save ( const SigmaI & SigI )

virtual

Regression value store.

Stores $t$ -statistic values for each trait and pre-selected (i.e. whether or not currently in the model) regression coefficient in a variable to subsequently save to a file by dump().

Parameters

[in] SigmaI& data inverse-covariance

Reimplemented from BetaGrpFt.

Member Data Documentation

◆ _selB

gsl_matrix* BetaGrpBVSR::_selB

protected

Matrix of selected predictors.

Contains only the predictors currently in the model.

The documentation for this class was generated from the following files:

Public Member Functions

Protected Member Functions

Protected Attributes

Friends

Detailed Description

Constructor & Destructor Documentation

◆ BetaGrpBVSR() [1/4]

◆ BetaGrpBVSR() [2/4]

◆ BetaGrpBVSR() [3/4]

◆ BetaGrpBVSR() [4/4]

Member Function Documentation

◆ _ldToss()

◆ _MGkernel() [1/3]

◆ _MGkernel() [2/3]

◆ _MGkernel() [3/3]

◆ save()

Member Data Documentation

◆ _selB