PeptDeepMS2Inference Class Reference

Inference engine for PeptDeep MS2 Fragment Intensity prediction. More...

#include <OpenMS/ML/PEPTDEEP/PeptDeepMS2Inference.h>

Public Types
enum	Channel : size_t { B_Z1 = 0 , B_Z2 = 1 , Y_Z1 = 2 , Y_Z2 = 3 }
	The channels of a cleavage position: [b_z1, b_z2, y_z1, y_z2, b_modloss_z1, b_modloss_z2, y_modloss_z1, y_modloss_z2]. More...

Public Member Functions
	PeptDeepMS2Inference (const std::string &model_path, int threads=4, size_t batch_size=500)
	Constructor initializes the ONNX session with the specified model graph.
	~PeptDeepMS2Inference ()
	Destructor.
std::vector< std::vector< float > >	predictMS2 (const std::vector< std::string > &peptides, const std::vector< float > &charges, const std::vector< float > &nces, const std::vector< int64_t > &instrument_indices)
	Predicts MS2 fragment intensities for a batch of peptide sequences.

Static Public Member Functions
static size_t	ionIndex (size_t length, char type, size_t ordinal, int charge, size_t channels=ION_CHANNELS)
	The index into a prediction of predictMS2() of the b or y ion of a peptide.

Static Public Attributes
static constexpr size_t	ION_CHANNELS = 8
	The values per cleavage position of a prediction of predictMS2(), in this order (Channel)
static constexpr size_t	NO_ION = static_cast<size_t>(-1)
	ionIndex() of an ion the prediction does not have

Private Attributes
ONNXPredictorBase	model_
size_t	batch_size_

Detailed Description

Inference engine for PeptDeep MS2 Fragment Intensity prediction.

This class parses and tokenizes raw peptide string sequences into structural integer arrays and evaluates them through the ONNX Runtime backend to predict fragment intensities.

Member Enumeration Documentation

Channel

enum Channel : size_t

The channels of a cleavage position: [b_z1, b_z2, y_z1, y_z2, b_modloss_z1, b_modloss_z2, y_modloss_z1, y_modloss_z2].

Enumerator
B_Z1
B_Z2
Y_Z1
Y_Z2

Constructor & Destructor Documentation

PeptDeepMS2Inference()

PeptDeepMS2Inference ( const std::string &  model_path, int  threads = 4, size_t  batch_size = 500  )

explicit

Constructor initializes the ONNX session with the specified model graph.

Parameters

model_path	Absolute or relative path to the PeptDeep MS2 ONNX model file.
threads	The model runs (of at most batch_size peptides of one length each) that run at once, each on one thread (default 4).
batch_size	Maximum number of peptides to process in a single ONNX run (default 500).

Exceptions

Exception::BaseException

if the model file cannot be found or loaded by the runtime.

~PeptDeepMS2Inference()

~PeptDeepMS2Inference

(

)

Destructor.

Member Function Documentation

ionIndex()

static size_t ionIndex ( size_t  length, char  type, size_t  ordinal, int  charge, size_t  channels = ION_CHANNELS  )

static

The index into a prediction of predictMS2() of the b or y ion of a peptide.

Cleavage position pos (0-based) carries b(pos + 1) and y(length - 1 - pos), with channels values each.

Parameters

length	The peptide's length (residues)
type	'b' or 'y'
ordinal	The ion's ordinal, 1 to length - 1
charge	1 or 2
channels	The values per cleavage position (ION_CHANNELS, unless a model has another layout)

Returns

The index, NO_ION for another ion type, ordinal or charge

predictMS2()

std::vector< std::vector< float > > predictMS2	(	const std::vector< std::string > &	peptides,
		const std::vector< float > &	charges,
		const std::vector< float > &	nces,
		const std::vector< int64_t > &	instrument_indices
	)

Predicts MS2 fragment intensities for a batch of peptide sequences.

Parameters

peptides	A vector of peptide strings. Supports OpenMS AASequence modification notation (e.g., "PEPTIDEK", "M(Oxidation)PEP").
charges	A vector of precursor charge states (must match peptides size).
nces	A vector of normalized collision energies (must match peptides size).
instrument_indices	One peptdeep instrument index per peptide. Derive these with ML::instrumentIndex() rather than writing literals: the encoding is positional and belongs to peptdeep (ML::ALPHAPEPTDEEP_INSTRUMENTS), where 0 is QE and 1 is Lumos. Indices outside [0, ML::PEPTDEEP_MAX_INSTRUMENT_NUM) are rejected, because the model's OneHot would otherwise answer them with an all-off row.

Returns

A vector of flattened fragment intensity arrays, one for each peptide. Each cleavage position contains eight contiguous AlphaPeptDeep channels in this order: [b_z1, b_z2, y_z1, y_z2, b_modloss_z1, b_modloss_z2, y_modloss_z1, y_modloss_z2].

Exceptions

Exception::IllegalArgument

if peptides is empty, size constraints fail, or a sequence is chemically invalid.

Member Data Documentation

batch_size_

size_t batch_size_

private

ION_CHANNELS

constexpr size_t ION_CHANNELS = 8

staticconstexpr

The values per cleavage position of a prediction of predictMS2(), in this order (Channel)

model_

ONNXPredictorBase model_

private

NO_ION

constexpr size_t NO_ION = static_cast<size_t>(-1)

staticconstexpr

ionIndex() of an ion the prediction does not have