MessagePasserFactory< Label > Class Template Reference

#include <OpenMS/ANALYSIS/ID/MessagePasserFactory.h>

Collaboration diagram for MessagePasserFactory< Label >:

Public Member Functions
evergreen::TableDependency< Label >	createProteinFactor (Label id, int nrMissingPeps=0)
	Protein Factor initialized with model prior (missing peps are experimental)
evergreen::TableDependency< Label >	createProteinFactor (Label id, double prior, int nrMissingPeps=0)
	Protein Factor initialized with user prior (missing peps are experimental)
evergreen::TableDependency< Label >	createPeptideEvidenceFactor (Label id, double prob)
evergreen::TableDependency< Label >	createRegularizingSumEvidenceFactor (size_t nr_parents, Label id, Label pep_id)
evergreen::TableDependency< Label >	createSumEvidenceFactor (size_t nr_parents, Label id, Label pep_id)
evergreen::TableDependency< Label >	createSumFactor (size_t nr_parents, Label nId)
evergreen::TableDependency< Label >	createReplicateFactor (Label seqId, Label repId)
evergreen::TableDependency< Label >	createChargeFactor (Label repId, Label chargeId, int chg)
evergreen::AdditiveDependency< Label >	createPeptideProbabilisticAdderFactor (const std::set< Label > &parentProteinIDs, Label nId)
	To sum up distributions for the number of parent proteins of a peptide with convolution trees.
evergreen::AdditiveDependency< Label >	createPeptideProbabilisticAdderFactor (const std::vector< Label > &parentProteinIDs, Label nId)
	To sum up distributions for the number of parent proteins of a peptide with convolution trees.
evergreen::PseudoAdditiveDependency< Label >	createBFPeptideProbabilisticAdderFactor (const std::set< Label > &parentProteinIDs, Label nId, const std::vector< evergreen::TableDependency< Label > > &deps)
	To sum up distributions for the number of parent proteins of a peptide brute-force.
	MessagePasserFactory (double alpha, double beta, double gamma, double p, double pep_prior)
	Constructor.
void	fillVectorsOfMessagePassers (const std::vector< Label > &protIDs, const std::vector< std::vector< Label > > &parentsOfPeps, const std::vector< double > &pepEvidences, evergreen::InferenceGraphBuilder< Label > &igb)
	Works on a vector of protein indices (potentially not consecutive)

Private Member Functions
double	notConditionalGivenSum (unsigned long summ)

Private Attributes
double	alpha_
	the model parameters
double	beta_
double	gamma_
double	p_
double	pepPrior_
std::map< int, double >	chgLLhoods = {{1, 0.7}, {2, 0.9}, {3, 0.7}, {4, 0.5}, {5, 0.5}}

Detailed Description

template<typename Label>
class OpenMS::Internal::MessagePasserFactory< Label >

Produces MessagePassers (nodes in a factor graph = bayesian network) for use with Evergreen library, based on a parameterization of the Protein-Peptide Bayesian network. Those MessagePassers can be tables or convolution trees. Labels are used to associate the variables they are working on. They can be integers (for speed) or strings (for readability/debug)

Constructor & Destructor Documentation

MessagePasserFactory()

template<typename Label >

MessagePasserFactory	(	double	alpha,
		double	beta,
		double	gamma,
		double	p,
		double	pep_prior
	)

Constructor.

Parameters

[in]	alpha	Peptide emission probability
[in]	beta	Spurious peptide emission probability
[in]	gamma	Protein prior
[in]	p	Marginalization norm
[in]	pep_prior	Peptide prior (defines at which evidence probability, additional evidence is beneficial)

Member Function Documentation

createBFPeptideProbabilisticAdderFactor()

template<typename Label >

evergreen::PseudoAdditiveDependency< Label > createBFPeptideProbabilisticAdderFactor	(	const std::set< Label > &	parentProteinIDs,
		Label	nId,
		const std::vector< evergreen::TableDependency< Label > > &	deps
	)

To sum up distributions for the number of parent proteins of a peptide brute-force.

createChargeFactor()

template<typename Label >

evergreen::TableDependency< Label > createChargeFactor	(	Label	repId,
		Label	chargeId,
		int	chg
	)

createPeptideEvidenceFactor()

template<typename Label >

evergreen::TableDependency< Label > createPeptideEvidenceFactor	(	Label	id,
		double	prob
	)

Peptide Factor initialized with:

Parameters

[in]	id	ID for the LabeledPMF
[in]	prob	peptide evidence probability

createPeptideProbabilisticAdderFactor() [1/2]

template<typename Label >

evergreen::AdditiveDependency< Label > createPeptideProbabilisticAdderFactor	(	const std::set< Label > &	parentProteinIDs,
		Label	nId
	)

To sum up distributions for the number of parent proteins of a peptide with convolution trees.

createPeptideProbabilisticAdderFactor() [2/2]

template<typename Label >

evergreen::AdditiveDependency< Label > createPeptideProbabilisticAdderFactor	(	const std::vector< Label > &	parentProteinIDs,
		Label	nId
	)

To sum up distributions for the number of parent proteins of a peptide with convolution trees.

createProteinFactor() [1/2]

template<typename Label >

evergreen::TableDependency< Label > createProteinFactor	(	Label	id,
		double	prior,
		int	nrMissingPeps = 0
	)

Protein Factor initialized with user prior (missing peps are experimental)

createProteinFactor() [2/2]

template<typename Label >

evergreen::TableDependency< Label > createProteinFactor	(	Label	id,
		int	nrMissingPeps = 0
	)

Protein Factor initialized with model prior (missing peps are experimental)

createRegularizingSumEvidenceFactor()

template<typename Label >

evergreen::TableDependency< Label > createRegularizingSumEvidenceFactor	(	size_t	nr_parents,
		Label	id,
		Label	pep_id
	)

Conditional probability table of peptide given number of parent proteins, based on model params. Additionally regularizes on the amount of parent proteins.

Parameters

[in]	nr_parents	(maximum) number of parent proteins
[in]	id	ID for the LabeledPMF
[in]	pep_id	ID for the LabeledPMF

createReplicateFactor()

template<typename Label >

evergreen::TableDependency< Label > createReplicateFactor	(	Label	seqId,
		Label	repId
	)

createSumEvidenceFactor()

template<typename Label >

evergreen::TableDependency< Label > createSumEvidenceFactor	(	size_t	nr_parents,
		Label	id,
		Label	pep_id
	)

Conditional probability table of peptide given number of parent proteins, based on model params.

Parameters

[in]	nr_parents	(maximum) number of parent proteins
[in]	id	ID for the LabeledPMF
[in]	pep_id	ID for the LabeledPMF

createSumFactor()

template<typename Label >

evergreen::TableDependency< Label > createSumFactor	(	size_t	nr_parents,
		Label	nId
	)

fillVectorsOfMessagePassers()

template<typename Label >

void fillVectorsOfMessagePassers	(	const std::vector< Label > &	protIDs,
		const std::vector< std::vector< Label > > &	parentsOfPeps,
		const std::vector< double > &	pepEvidences,
		evergreen::InferenceGraphBuilder< Label > &	igb
	)

Works on a vector of protein indices (potentially not consecutive)

notConditionalGivenSum()

template<typename Label >

double notConditionalGivenSum ( unsigned long  summ )

inlineprivate

to fill the noisy-OR table for a peptide given parent proteins TODO pre-compute for like a hundred parent proteins TODO introduce special case for alpha or beta = 1. The log formula does not work otherwise.

References MessagePasserFactory< Label >::alpha_, and MessagePasserFactory< Label >::beta_.

Member Data Documentation

alpha_

template<typename Label >

double alpha_

private

the model parameters

Referenced by MessagePasserFactory< Label >::notConditionalGivenSum().

beta_

template<typename Label >

double beta_

private

Referenced by MessagePasserFactory< Label >::notConditionalGivenSum().

chgLLhoods

template<typename Label >

std::map<int, double> chgLLhoods = {{1, 0.7}, {2, 0.9}, {3, 0.7}, {4, 0.5}, {5, 0.5}}

private

Likelihoods for the charge states given presence of the peptide sequence (

Todo:

could be calculated from IDPEP if we do per charge state fitting) or empirically estimated from the input PSMs

gamma_

template<typename Label >

double gamma_

private

p_

template<typename Label >

double p_

private

pepPrior_

template<typename Label >

double pepPrior_

private