pulsatrix 1.1.0
Loading...
Searching...
No Matches
pulsatrix::EncoderTarget Struct Reference

What an explanation explains. Residues are numbered from 0 here. More...

#include <protein_explanations.hpp>

Collaboration diagram for pulsatrix::EncoderTarget:

Public Types

enum class  Kind { MaskedToken , Mutation , ProteinHead , ResidueHead }
 

Public Member Functions

std::string describe (std::string_view sequence) const
 For labels: "K11 masked: R", "K11R log-odds", "protein head", "head at K11".
 

Static Public Member Functions

static EncoderTarget MaskedToken (int64_t residue, char token)
 
static EncoderTarget ForMutation (const Mutation &m, int64_t offset=1)
 The mutation m; its position is numbered from offset, as variant tables do.
 
static EncoderTarget ProteinHead (LinearHead head)
 
static EncoderTarget ResidueHead (LinearHead head, int64_t residue)
 

Public Attributes

Kind kind = Kind::MaskedToken
 
int64_t residue = 0
 
char token = 0
 
char wild_type = 0
 
LinearHead head
 

Detailed Description

What an explanation explains. Residues are numbered from 0 here.

Member Enumeration Documentation

◆ Kind

enum class pulsatrix::EncoderTarget::Kind
strong
Enumerator
MaskedToken 

The logit of token at residue, with that residue masked.

Mutation 

logit(token) - logit(wild_type) at residue, masked: a mutation's log-odds, the masked-marginal score (VariantScorer) before log-softmax, which cancels.

ProteinHead 

A per-protein head on the mean of the residues' representations (<cls> and <eos> left out, as ESM embeddings are usually pooled), unmasked.

ResidueHead 

A per-residue head on residue's representation, unmasked.

Member Function Documentation

◆ describe()

std::string pulsatrix::EncoderTarget::describe ( std::string_view  sequence) const

For labels: "K11 masked: R", "K11R log-odds", "protein head", "head at K11".

◆ ForMutation()

static EncoderTarget pulsatrix::EncoderTarget::ForMutation ( const Mutation &  m,
int64_t  offset = 1 
)
static

The mutation m; its position is numbered from offset, as variant tables do.

◆ MaskedToken()

static EncoderTarget pulsatrix::EncoderTarget::MaskedToken ( int64_t  residue,
char  token 
)
static

◆ ProteinHead()

static EncoderTarget pulsatrix::EncoderTarget::ProteinHead ( LinearHead  head)
static

◆ ResidueHead()

static EncoderTarget pulsatrix::EncoderTarget::ResidueHead ( LinearHead  head,
int64_t  residue 
)
static

Member Data Documentation

◆ head

LinearHead pulsatrix::EncoderTarget::head

◆ kind

Kind pulsatrix::EncoderTarget::kind = Kind::MaskedToken

◆ residue

int64_t pulsatrix::EncoderTarget::residue = 0

◆ token

char pulsatrix::EncoderTarget::token = 0

◆ wild_type

char pulsatrix::EncoderTarget::wild_type = 0

The documentation for this struct was generated from the following file: