Create and apply models of a mass recalibration function. More...

#include <OpenMS/FILTERING/CALIBRATION/MZTrafoModel.h>

Collaboration diagram for MZTrafoModel:

Classes
struct	RTLess
	Comparator by position. As this class has dimension 1, this is basically an alias for MZLess. More...

Public Types
enum	MODELTYPE { LINEAR , LINEAR_WEIGHTED , QUADRATIC , QUADRATIC_WEIGHTED , SIZE_OF_MODELTYPE }

Public Member Functions
	MZTrafoModel ()
	Default constructor. More...

	MZTrafoModel (bool ppm_model)
	Default constructor. More...

bool	isTrained () const
	Does the model have coefficients (i.e. was trained successfully). More...

double	getRT () const
	Get RT associated with the model (training region) More...

double	predict (double mz) const
	Apply the model to an uncalibrated m/z value. More...

bool	train (const CalibrationData &cd, MODELTYPE md, bool use_RANSAC, double rt_left=-std::numeric_limits< double >::max(), double rt_right=std::numeric_limits< double >::max())
	Train a model using calibrant data. More...

bool	train (std::vector< double > error_mz, std::vector< double > theo_mz, std::vector< double > weights, MODELTYPE md, bool use_RANSAC)
	Train a model using calibrant data. More...

void	getCoefficients (double &intercept, double &slope, double &power)
	Get model coefficients. More...

void	setCoefficients (const MZTrafoModel &rhs)
	Copy model coefficients from another model. More...

void	setCoefficients (double intercept, double slope, double power)
	Manually set model coefficients. More...

String	toString () const
	String representation of the model parameters. More...

Static Public Member Functions
static MODELTYPE	nameToEnum (const std::string &name)
	Convert string to enum. More...

static const std::string &	enumToName (MODELTYPE mt)
	Convert enum to string. More...

static void	setRANSACParams (const Math::RANSACParam &p)
	Set the global (program wide) parameters for RANSAC. More...

static void	setRANSACSeed (int seed)
	Set RANSAC seed. More...

static void	setCoefficientLimits (double offset, double scale, double power)
	Set coefficient boundaries for which the model coefficient must not exceed to be considered a valid model. More...

static bool	isValidModel (const MZTrafoModel &trafo)
	Predicate to decide if the model has valid parameters, i.e. coefficients. More...

static Size	findNearest (const std::vector< MZTrafoModel > &tms, double rt)
	Binary search for the model nearest to a specific RT. More...

Static Public Attributes
static const std::string	names_of_modeltype []
	strings corresponding to enum MODELTYPE More...

Private Attributes
std::vector< double >	coeff_
	Model coefficients (for both linear and quadratic models), estimated from the data. More...

bool	use_ppm_
	during training, model is build on absolute or relative(ppm) predictions. predict(), i.e. applying the model, requires this information too More...

double	rt_
	retention time associated to the model (i.e. where the calibrant data was taken from) More...

Static Private Attributes
static Math::RANSACParam *	ransac_params_
	global pointer, init to NULL at startup; set class-global RANSAC params More...

static int	ransac_seed_
	seed used for all RANSAC invocations More...

static double	limit_offset_
	acceptable boundary for the estimated offset; if estimated offset is larger (absolute) the model does not validate (isValidModel()) More...

static double	limit_scale_
	acceptable boundary for the estimated scale; if estimated scale is larger (absolute) the model does not validate (isValidModel()) More...

static double	limit_power_
	acceptable boundary for the estimated power; if estimated power is larger (absolute) the model does not validate (isValidModel()) More...

Detailed Description

Create and apply models of a mass recalibration function.

The input is a list of calibration points (ideally spanning a wide m/z range to prevent extrapolation when applying to model).

Models (LINEAR, LINEAR_WEIGHTED, QUADRATIC, QUADRATIC_WEIGHTED) can be trained using CalData points (or a subset of them). Calibration points can have different retention time points, and a model should be build such that it captures the local (in time) decalibration of the instrument, i.e. choose appropriate time windows along RT to calibrate the spectra in this RT region. From the available calibrant data, a model is build. Later, any uncalibrated m/z value can be fed to the model, to obtain a calibrated m/z.

The input domain can either be absolute mass differences in [Th], or relative differences in [ppm]. The models are build based on this input.

Outlier detection before model building via the RANSAC algorithm is supported for LINEAR and QUADRATIC models.

Member Enumeration Documentation

◆ MODELTYPE

enum MODELTYPE

Enumerator
LINEAR
LINEAR_WEIGHTED
QUADRATIC
QUADRATIC_WEIGHTED
SIZE_OF_MODELTYPE

Constructor & Destructor Documentation

◆ MZTrafoModel() [1/2]

MZTrafoModel ( )

Default constructor.

◆ MZTrafoModel() [2/2]

MZTrafoModel ( bool ppm_model )

Default constructor.

If you have external coefficients, use this constructor and the setCoefficients() method to build a 'manual' model. Afterwards, use applyTransformation() or predict() to calibrate your data. If you call train(), the ppm-setting will be overwritten, depending on the type of training data.

Parameters

ppm_model Are the coefficients derived from ppm calibration data, or from absolute deltas?

Member Function Documentation

◆ enumToName()

static const std::string& enumToName ( MODELTYPE mt )

static

Convert enum to string.

Parameters

mt	The enum value

Returns: Stringified version

◆ findNearest()

static Size findNearest	(	const std::vector< MZTrafoModel > &	tms,
		double	rt
	)

static

Binary search for the model nearest to a specific RT.

Parameters

tms	Vector of models, sorted by RT
rt	The target retention time

Returns: Returns the index into 'tms' with the closest RT.

Note: Make sure the vector is sorted with respect to RT! Otherwise the result is undefined.

Exceptions

Exception::Precondition is thrown if the vector is empty (not only in debug mode)

◆ getCoefficients()

void getCoefficients	(	double &	intercept,
		double &	slope,
		double &	power
	)

Get model coefficients.

Parameters will be filled with internal model parameters. The model must be trained before; Exception is thrown otherwise!

Parameters

intercept	The intercept
slope	The slope
power	The coefficient for x*x (will be 0 for linear models)

Exceptions

Exception::Precondition if model is not trained yet

◆ getRT()

double getRT ( ) const

Get RT associated with the model (training region)

◆ isTrained()

bool isTrained ( ) const

Does the model have coefficients (i.e. was trained successfully).

Having coefficients does not mean its valid (see isValidModel(); since coeffs might be too large).

◆ isValidModel()

static bool isValidModel ( const MZTrafoModel & trafo )

static

Predicate to decide if the model has valid parameters, i.e. coefficients.

If the model coefficients are empty, no model was trained yet (or unsuccessful), causing a return value of 'false'.

Also, if the model has coefficients, we check if they are within the acceptable boundaries (if boundaries were given via setCoeffientLimits()).

◆ nameToEnum()

static MODELTYPE nameToEnum ( const std::string & name )

static

Convert string to enum.

Returns 'SIZE_OF_MODELTYPE' if string is unknown.

Parameters

name	A string from names_of_modeltype[].

Returns: The corresponding enum value.

◆ predict()

double predict ( double mz ) const

Apply the model to an uncalibrated m/z value.

Make sure the model was trained (train()) and is valid (isValidModel()) before calling this function!

Applies the function y = intercept + slope*mz + power*mz^2 and returns y.

Parameters

mz	The uncalibrated m/z value

Returns: The calibrated m/z value

◆ setCoefficientLimits()

static void setCoefficientLimits	(	double	offset,
		double	scale,
		double	power
	)

static

Set coefficient boundaries for which the model coefficient must not exceed to be considered a valid model.

Use std::numeric_limits<double>::max() for no limit (default). If isValidModel() is called these limits are checked. Negative input run through fabs() to get positive values (since comparison is done in absolute terms).

◆ setCoefficients() [1/2]

void setCoefficients ( const MZTrafoModel & rhs )

Copy model coefficients from another model.

◆ setCoefficients() [2/2]

void setCoefficients	(	double	intercept,
		double	slope,
		double	power
	)

Manually set model coefficients.

Can be used instead of train(), so manually set coefficients. It must be exactly three values. If you want a linear model, set 'power' to zero. If you want a constant model, set slope to zero in addition.

Parameters

intercept	The offset
slope	The slope
power	The x*x coefficient (for quadratic models)

◆ setRANSACParams()

static void setRANSACParams ( const Math::RANSACParam & p )

static

Set the global (program wide) parameters for RANSAC.

This is not done via member, to keep a small memory footprint since hundreds of MZTrafoModels are expected to be build at the same time and the RANSAC params should be identical for all of them.

Parameters

p	RANSAC params

◆ setRANSACSeed()

static void setRANSACSeed ( int seed )

static

Set RANSAC seed.

◆ toString()

String toString ( ) const

String representation of the model parameters.

Empty if model is not trained.

◆ train() [1/2]

bool train	(	const CalibrationData &	cd,
		MODELTYPE	md,
		bool	use_RANSAC,
		double	rt_left = `-std::numeric_limits< double >::max()`,
		double	rt_right = `std::numeric_limits< double >::max()`
	)

Train a model using calibrant data.

If the CalibrationData were created using peak groups (usually corresponding to mass traces), the median for each group is used as a group representative. This is more robust, and reduces the number of data points drastically, i.e. one value per group.

Internally, these steps take place:

apply RT filter
[compute median per group] (only if groups were given in 'cd')
set Model's rt position
call train() (see overloaded method)

Parameters

cd	List of calibrants
md	Type of model (linear, quadratic, ...)
use_RANSAC	Remove outliers before computing the model?
rt_left	Filter 'cd' by RT; all calibrants with RT < 'rt_left' are removed
rt_right	Filter 'cd' by RT; all calibrants with RT > 'rt_right' are removed

Returns: True if model was build, false otherwise

◆ train() [2/2]

bool train	(	std::vector< double >	error_mz,
		std::vector< double >	theo_mz,
		std::vector< double >	weights,
		MODELTYPE	md,
		bool	use_RANSAC
	)

Train a model using calibrant data.

Given theoretical and observed mass values (and corresponding weights), a model (linear, quadratic, ...) is build. Outlier removal is applied before. The 'obs_mz' can be either given as absolute masses in [Th] or relative deviations in [ppm]. The MZTrafoModel must be constructed accordingly (see constructor). This has no influence on the model building itself, but rather on how 'predict()' works internally.

Outlier detection before model building via the RANSAC algorithm is supported for LINEAR and QUADRATIC models.

Internally, these steps take place:

[apply RANSAC] (depending on 'use_RANSAC')
build model and store its parameters internally

Parameters

error_mz	Observed Mass error (in ppm or Th)
theo_mz	Theoretical m/z values, corresponding to 'error_mz'
weights	For weighted models only: weight of calibrants; ignored otherwise
md	Type of model (linear, quadratic, ...)
use_RANSAC	Remove outliers before computing the model?

Returns: True if model was build, false otherwise

Member Data Documentation

◆ coeff_

std::vector<double> coeff_

private

Model coefficients (for both linear and quadratic models), estimated from the data.

◆ limit_offset_

double limit_offset_

staticprivate

acceptable boundary for the estimated offset; if estimated offset is larger (absolute) the model does not validate (isValidModel())

◆ limit_power_

double limit_power_

staticprivate

acceptable boundary for the estimated power; if estimated power is larger (absolute) the model does not validate (isValidModel())

◆ limit_scale_

double limit_scale_

staticprivate

acceptable boundary for the estimated scale; if estimated scale is larger (absolute) the model does not validate (isValidModel())

◆ names_of_modeltype

const std::string names_of_modeltype[]

static

strings corresponding to enum MODELTYPE

◆ ransac_params_

Math::RANSACParam* ransac_params_

staticprivate

global pointer, init to NULL at startup; set class-global RANSAC params

◆ ransac_seed_

int ransac_seed_

staticprivate

seed used for all RANSAC invocations

◆ rt_

double rt_

private

retention time associated to the model (i.e. where the calibrant data was taken from)

Referenced by MZTrafoModel::RTLess::operator()().

◆ use_ppm_

bool use_ppm_

private

during training, model is build on absolute or relative(ppm) predictions. predict(), i.e. applying the model, requires this information too

Classes

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Private Attributes

Static Private Attributes

Detailed Description

Member Enumeration Documentation

◆ MODELTYPE

Constructor & Destructor Documentation

◆ MZTrafoModel() [1/2]

◆ MZTrafoModel() [2/2]

Member Function Documentation

◆ enumToName()

◆ findNearest()

◆ getCoefficients()

◆ getRT()

◆ isTrained()

◆ isValidModel()

◆ nameToEnum()

◆ predict()

◆ setCoefficientLimits()

◆ setCoefficients() [1/2]

◆ setCoefficients() [2/2]

◆ setRANSACParams()

◆ setRANSACSeed()

◆ toString()

◆ train() [1/2]

◆ train() [2/2]

Member Data Documentation

◆ coeff_

◆ limit_offset_

◆ limit_power_

◆ limit_scale_

◆ names_of_modeltype

◆ ransac_params_

◆ ransac_seed_

◆ rt_

◆ use_ppm_