Table 2.

PLI prediction methods as regression tasks based on the ML framework in recent years^a.

Toolb	Date	Input protein features	Input compound features	Protein feature extractor	Compound feature extractor	Methods
SimBoost [26]	04/2017	Target similarity	Drug similarity	–	–	Gradient boosting tree model
ACNN [83]	2017	Atomic coordinates	Atomic coordinates	Atomic convolution layer	Atomic convolution layer	Atomic fully connected layer
DeepDTA [84]	09/2018	Label encoding	Label encoding	CNN blocks	CNN blocks	Fully connected layer
DeepAffinity [46]	02/2019	Structural property sequence representation	Structural property sequence representation	Seq2seq autoencoders	Seq2seq autoencoders	Unified RNN-CNN
WideDTA [85]	02/2019	Textual information	Textual information	CNN blocks	CNN blocks	Fully connected layers
GraphDTA [86]	06/2019	One-hot encoding	Molecular graph	Convolutional layers	4 graph neural network variants	Fully connected layers
RFScore [17]	08/2019	36 intermolecular features	36 intermolecular features	–	–	Random forest
AttentionDTA [36]	11/2019	Label encoding	Label encoding	CNN block	CNN block	Attention block- fully connected layers
Taba [87]	01/2020	The average distance between pairs of atoms	The average distance between pairs of atoms	–	–	Machine-learning model
GAT_GCN [88]	04/2020	Peptide frequency	Graph structure	CNN	GCN	Fully connected layers
SAnDReS [89]	05/2020	Docking scores	Docking scores	–	–	Machine-learning model
DeepCDA [90]	05/2020	N-gram embedding	SMILES sequence	CNN-LSTM-Two-sided attention mechanism	CNN-LSTM-Two-sided attention mechanism	Fully connected layers
DGraphDTA [91]	06/2020	Protein graph	Molecular graph	GNN	GNN	Fully connected layers
JoVA [92]	08/2020	Multiple unimodal representations	Multiple unimodal representations	Joint view attention module	Joint view attention module	Prediction model
Fusion [93]	11/2020	Atomic representation	Atomic representation	CNNs	SG-GCNs	Fully connected layers
DeepGS [44]	2020	Symbolic sequences	Molecular structure	Prot2Vec-CNN-BiGRU blocks	Smi2Vec-CNN-BiGRU blocks	Fully connected layer
DeepDTAF [94]	01/2021	Sequence, structural property information	SMILES string	Dilated/traditional convolution layers	Dilated convolution layers	Fully connected layers
GanDTI [37] (classification and regression)	03/2021	Protein sequences	Molecule fingerprints-adjacency matrix	Attention module	Residual graph neural network	MLP
Multi-PLI [38] (classification and regression)	04/2021	One-hot vectors	One-hot vectors	CNN blocks	CNN blocks	Fully connected layers
ML-DTI [95]	04/2021	Protein sequences	SMILES string	CNN block (mutual learning)	CNN block (mutual learning)	Linear transformation layers
DEELIG [47]	06/2021	Atomic level-structural information-sequences	Physical properties-fingerprints	CNN	Fully connected layers	Fully connected layers
GEFA [55]	07/2021	Sequence embedding features	Graph representation	GCN	GCN	Linear layers
SAG-DTA [96]	08/2021	Label encoding	Molecular graph	CNN	Graph convolutional layer-SAGPooling layer	Fully connected layers
Tanoori et al. [97]	08/2021	SW sequence similarity	CS similarity	–	–	GBM
EmbedDTI [56]	11/2021	Amino acids	Structural information	CNN	Attention-GCNs	Fully connected layers
DeepPLA [45]	12/2021	Protein sequences (ProSE)	SMILES strings (Mol2Vec)	Head CNN modules-ResNet-based CNN module	Head CNN modules-ResNet-based CNN module	BiLSTM module-MLP module
DeepGLSTM [98]	01/2022	Amino acids	Adjacency representation	BiLSTM	GCN	Fully connected layers
MGraphDTA [99]	01/2022	Integers	Graph structure	Multiscale convolutional neural network	GNN	MLP
FusionDTA [100]	01/2022	word embeddings	SMILES strings	BiLSTM	BiLSTM	Multi- head linear attention blocks/Fully connected layer
HoTS [39] (classification and regression)	02/2022	Protein sequences	Morgan/circular fingerprints	Transformer blocks	Transformer blocks	Fully connected layers
ELECTRA-DTA [101]	03/2022	Protein sequences	SMILES string	Squeeze-and-excitation convolutional neural network blocks	Squeeze-and-excitation convolutional neural network blocks	Fully connected layers

Note: “-” in the table indicates that there is no such information in the corresponding article.

^aAbbreviations: CNN – convolutional neural network; GNN – graph neural network; GCNs – graph convolutional networks; LSTM – long short-term memory; SG-CNNs – spatial graph neural networks; BiGRU – bidirectional gate recurrent unit; MLP – multilayer perceptron; GCN – graph convolutional network; SW – Smith-Waterman; CS – chemical structure; GBM – gradient boosting machine; BiLSTM – bidirectional long short-term memory;

^bURL addresses for the listed tools: SimBoost – https://zenodo.org/record/164436; ACNN – https://github.com/deepchem/deepchem; DeepDTA – https://github.com/hkmztrk/DeepDTA; DeepAffinity – https://github.com/Shen-Lab/DeepAffinity; GraphDTA – https://github.com/thinng/GraphDTA; Taba – https://github.com/azevedolab/taba; SAnDReS – https://github.com/azevedolab/sandres; DeepCDA – https://github.com/LBBSoft/DeepCDA; Fusion – https://github.com/llnl/fast; DeepGS – https://github.com/jacklin18/DeepGS; DeepDTAF – https://github.com/KailiWang1/DeepDTAF; GanDTI – https://github.com/shuyu-wang/GanDTI; Multi-PLI – https://github.com/Deshan-Zhou/Multi-PLI; ML-DTI – https://github.com/guaguabujianle/ML-DTI.git; DEELIG – https://github.com/asadahmedtech/DEELIG; GEFA – https://github.com/ngminhtri0394/GEFA; EmbedDTI – https://github.com/Aurorayuan/EmbedDTI; DeepPLA – https://github.com/David-BominWei/DeepPLA; DeepGLSTM – https://github.com/MLlab4CS/DeepGLSTM.git; MGraphDTA – https://github.com/guaguabujianle/MGraphDTA; FusionDTA – https://github.com/yuanweining/FusionDTA; HoTS – https:// github. com/ GIST- CSBL/ HoTS; ELECTRA-DTA – https://github.com/IILab-Resource/ELECTRA-DTA.