de.jstacs.sequenceScores.statisticalModels.trainable
Class TrainableStatisticalModelFactory

java.lang.Object
  de.jstacs.sequenceScores.statisticalModels.trainable.TrainableStatisticalModelFactory

public class TrainableStatisticalModelFactory
extends Object

This class allows to easily create some frequently used models. It offers only one way of creating each model and set some of the parameters to default values. If you like to set further models please check the constructors of the individual classes.

Author:

Jens Keilwagen

Constructor Summary

TrainableStatisticalModelFactory()

Method Summary

static BayesianNetworkTrainSM	createBayesianNetworkModel(AlphabetContainer con, int length, double ess, byte order) This method returns a Bayesian network model (BN) with user-specified order.
static HomogeneousMM	createHomogeneousMarkovModel(AlphabetContainer con, double ess, byte order) This method returns a homogeneous Markov model with user-specified order.
static FSDAGTrainSM	createInhomogeneousMarkovModel(AlphabetContainer con, int length, double ess, byte order) This method returns a inhomogeneous Markov model (IMM) with user-specified order.
static MixtureTrainSM	createMixtureModel(double[] hyper, TrainableStatisticalModel[] model) This method allows to create a MixtureTrainSM that allows to model a DataSet as a mixture of individual components.
static BayesianNetworkTrainSM	createPermutedMarkovModel(AlphabetContainer con, int length, double ess, byte order) This method returns a permuted Markov model (PMM) with user-specified order.
static FSDAGTrainSM	createPWM(AlphabetContainer con, int length, double ess) This method returns a position weight matrix (PWM).
static StrandTrainSM	createStrandModel(TrainableStatisticalModel model) This method allows to create a StrandTrainSM that allows to score binding sites on both strand of DNA.
static ZOOPSTrainSM	createZOOPS(TrainableStatisticalModel motif, TrainableStatisticalModel bg, double[] hyper, boolean trainOnlyMotifModel) This method allows to create a "zero or one occurrence per sequence" (ZOOPS) model that allows to discover binding sites in a DataSet.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

TrainableStatisticalModelFactory

public TrainableStatisticalModelFactory()

Method Detail

createPWM

public static FSDAGTrainSM createPWM(AlphabetContainer con,
                                     int length,
                                     double ess)
                              throws Exception

This method returns a position weight matrix (PWM). A PWM assumes that all positions of a sequence are statistically independent.

Parameters:

con - the AlphabetContainer of the PWM

length - the length of the PWM, i.e., the length of the sequences that can be modeled

ess - the equivalent sample size (ess) of the PWM, if 0 (zero) the model can be trained using maximum likelihood principle, otherwise it can be trained using the maximum a posteriori principle using the BDeu prior

Returns:

the PWM

Throws:

Exception - if the model can not be created correctly

createInhomogeneousMarkovModel

public static FSDAGTrainSM createInhomogeneousMarkovModel(AlphabetContainer con,
                                                          int length,
                                                          double ess,
                                                          byte order)
                                                   throws Exception

This method returns a inhomogeneous Markov model (IMM) with user-specified order.

Parameters:

con - the AlphabetContainer of the IMM

length - the length of the IMM, i.e., the length of the sequences that can be modeled

ess - the equivalent sample size (ess) of the IMM, if 0 (zero) the model can be trained using maximum likelihood principle, otherwise it can be trained using the maximum a posteriori principle using the BDeu prior

order - the order of the IMM, i.e., the number of directly preceding random variables (=positions) that might have an influence on the probability of outcome of a random variable (=position)

Returns:

the IMM

Throws:

Exception - if the model can not be created correctly

createPermutedMarkovModel

public static BayesianNetworkTrainSM createPermutedMarkovModel(AlphabetContainer con,
                                                               int length,
                                                               double ess,
                                                               byte order)
                                                        throws Exception

This method returns a permuted Markov model (PMM) with user-specified order. Permuted Markov models determine a permutation of the random variables (=position) and than build a inhomogeneous Markov model based on this permutation.

Parameters:

con - the AlphabetContainer of the PMM

length - the length of the PMM, i.e., the length of the sequences that can be modeled

ess - the equivalent sample size (ess) of the PMM, if 0 (zero) the model can be trained using maximum likelihood principle, otherwise it can be trained using the maximum a posteriori principle using the BDeu prior

order - the order of the PMM, i.e., the number of random variables (=positions) that might have an influence on the probability of outcome of a random variable (=position)

Returns:

the PMM

Throws:

Exception - if the model can not be created correctly

createBayesianNetworkModel

public static BayesianNetworkTrainSM createBayesianNetworkModel(AlphabetContainer con,
                                                                int length,
                                                                double ess,
                                                                byte order)
                                                         throws Exception

This method returns a Bayesian network model (BN) with user-specified order. Bayesian network determine a directed acyclic graph of the random variables (positions) and than learn the parameters of the distribution based on this network.

Parameters:

con - the AlphabetContainer of the BN

length - the length of the BN, i.e., the length of the sequences that can be modeled

ess - the equivalent sample size (ess) of the BN, if 0 (zero) the model can be trained using maximum likelihood principle, otherwise it can be trained using the maximum a posteriori principle using the BDeu prior

order - the order of the BN, i.e., the number of random variables (=positions) that might have an influence on the probability of outcome of a random variable (=position)

Returns:

the BN

Throws:

Exception - if the model can not be created correctly

createHomogeneousMarkovModel

public static HomogeneousMM createHomogeneousMarkovModel(AlphabetContainer con,
                                                         double ess,
                                                         byte order)
                                                  throws Exception

This method returns a homogeneous Markov model with user-specified order.

Parameters:

con - the AlphabetContainer of the model

ess - the equivalent sample size (ess) of the model, if 0 (zero) the model can be trained using maximum likelihood principle, otherwise it can be trained using the maximum a posteriori principle using the BDeu prior

order - the order of the model, i.e., the number of directly preceding random variables (=positions) that might have an influence on the probability of outcome of a random variable (=position)

Returns:

the homogeneous Markov model

Throws:

Exception - if the model can not be created correctly

createStrandModel

public static StrandTrainSM createStrandModel(TrainableStatisticalModel model)
                                       throws Exception

This method allows to create a StrandTrainSM that allows to score binding sites on both strand of DNA.

Parameters:

model - the internally used model

Returns:

the StrandTrainSM

Throws:

Exception - if the model can not be created correctly

createMixtureModel

public static MixtureTrainSM createMixtureModel(double[] hyper,
                                                TrainableStatisticalModel[] model)
                                         throws Exception

This method allows to create a MixtureTrainSM that allows to model a DataSet as a mixture of individual components.

Parameters:

hyper - the hyper parameters for the components (should be identical to the ESS of the components)

model - the internally used model

Returns:

the MixtureTrainSM

Throws:

Exception - if the model can not be created correctly

createZOOPS

public static ZOOPSTrainSM createZOOPS(TrainableStatisticalModel motif,
                                       TrainableStatisticalModel bg,
                                       double[] hyper,
                                       boolean trainOnlyMotifModel)
                                throws Exception

This method allows to create a "zero or one occurrence per sequence" (ZOOPS) model that allows to discover binding sites in a DataSet.

Parameters:

motif - the internally used model for the binding sites

bg - the internally used model for the flanking sequence

hyper - the hyper parameters for the components (should be identical to the ESS of the components)

trainOnlyMotifModel - a switch allowing to train either the motif model or both (motif and bg) models

Returns:

the ZOOPS model

Throws:

Exception - if the model can not be created correctly