Table 3 Application of machine learning technologies in neoantigen and immunogenicity prediction

From: Informing immunotherapy with multi-omics driven machine learning

Model

Task

ML Model

Encoding method

MHC class

Ref

NetMHC-4.0

Predict binding affinity

NN

BLOSUM

class I

99

NetMHCpan-4.0

Predict binding affinity

NN

BLOSUM

class I

100

MHCflurry

Predict binding affinity

A deep learning model includes locally connected

1D-CNN and FCNN

BLOSUM

class I

101

EDGE

Predict binding affinity

NN

One-hot

class I

102

MHCRoBERTa

Predict binding affinity

BPE

Byte pair

class I

103

Vang et al.

Predict binding affinity

CNN

Skip-gram

class I

104

ForestMHC

Predict binding affinity

RF

NA

class I

105

Anthem

Predict binding affinity

NB, XGBoost, LR, NN, SVM, DT, RF

BLOSUM

class I

106

Gartner et al.

Rank binding affinity

RF

NA

class I

107

NN-align

Predict binding affinity

NN

BLOSUM

class II

110

MixMHC2pred

Predict binding affinity

Linear regression

BLOSUM

class II

111

NeonMHC2

Predict binding affinity

CNN

One-hot

class II

112

NetMHCIIpan

Predict binding affinity

NN

BLOSUM

class II

113

MARIA

Predict binding affinity

LSTM

One-hot

class II

114

Neopepsee

Predict immunogenicity

LNB, GNB, RF, SVM

NA

class I

115

DeepNetBim

Predict immunogenicity

A deep learning model includes CNN and attention module

BLOSUM

class I

116

DeepHLApan

Predict immunogenicity

BiGRU + attention

One-hot

class I

117

Seq2Neo

Predict immunogenicity

CNN

One-hot

class I

118

TCIA

Predict cancer immunogenicity

RF

NA

class I, class II, non-classical

120

Besser et al.

Predict CD8 + T cell response

RF

NA

class I

121

iTTCA-Hybrid

Predict CD8 + T cell response

SVM, RF

NA

class I

123

DLpTCR

Predict TCR-pMHC interaction

A multimodal model based on FCNN, LeNet-5, ResNet-20

One-hot, PCA, PCP

class I

124

pMTnet

Predict TCR-pMHC interaction

AE + LSTM + NN

BLOSUM

class I

125

  1. NN neural network, BPE byte pair encoding, CNN convolutional neural network, FCNN fully connected neural network, RF random forest, NA not applicable, NB naïve Bayes, LR logistic regression, SVM support vector machines, DT decision tree, LSTM long short-term memory, LNB locally weighted naïve Bayes, GNB Gaussian naïve Bayes, BiGRU bidirectional Gated Recurrent Unit, pMHC peptide-MHC, PCA principal component analysis, PCP physicochemical properties, AE autoencoder.
Back to article page