Table 1.

Summary of tools: category, architecture, strength/uniqueness, and availability of the tools described in this article.

Category	Tool	Architecture	Strength	Code/Web Server (Last Accessed on 10 May 2021)
End-to-end structure prediction	AlQuarishi’s end-to-end model [14]	Recurrent geometric network (RGN)	Predicted novel folds without co-evolutionary data, it achieved state-of-the-art accuracy	https://github.com/aqlaboratory/rgn
	NEMO [76]	DL	First end-to-end Deep Learning-based approach	NA
	AlphaFold 2	Transformers (attention mechanism)	Evolutionary related sequences and MSA are fetched into transformers to accurately predict protein 3D structure	NA
Real-valued distance prediction	PDNET [19]	ResNet	A fully open-source and light framework for distance, contact, and distogram prediction	https://github.com/ba-lab/pdnet/
	GAN-based method [18]	GAN+ ResNet	One of the initial efforts to predict real-valued distance maps; GANs developed to predict real-valued distance maps	https://github.com/Wenze-Codebase/DistancePrediction-Protein-GAN.git http://structpred.life.tsinghua.edu.cn/continental.html [W]
	Xu’s method [62]	ResNet	Predicts not only real-valued distance but also mean and deviation of a distance for folding	NA
	REALDIST [17]	ResNet	Highly accurate distance prediction method focusing only on real-valued distance map predictions and distance-guided 3D modeling	https://github.com/ba-lab/realdist
	DeepDist [16]	ResNet	Predicts both distograms and real-valued distances and delivers high-accuracy distance maps	https://github.com/multicom-toolbox/deepdist
Distogram (Smaller Distance range) prediction	RaptorX [11,43]	ResNet	The original RaptorX method upgraded to predict distograms	http://raptorx.uchicago.edu/AbInitioFolding/ [W], https://github.com/j3xugit/RaptorX-Contact
	ProSPr [15]	ResNet	An open-source protein distance prediction network inspired from the AlphaFold implementation	https://github.com/dellacortelab/prospr
	trRosetta [56]	ResNet	A fully Tensorflow-based open-source implementation to predict distograms; demonstrated to outperform AlphaFold	https://github.com/gjoni/trRosetta https://yanglab.nankai.edu.cn/trRosetta/ [W]
	DeepH3 [74]	ResNet	It predicts inter-residue distances and orientation from antibody heavy and light chain sequences	https://github.com/Graylab/deepH3-distances-orientations
	AttentiveDist [61]	RestNet with Attention	It uses MSAs generated with different E-values to increase the co-evolutionary information provided to the model	https://github.com/kiharalab/AttentiveDist
	DISTEVAL [101]		A tool and web server for evaluating predicted real-values distances, distograms, and contacts	https://github.com/ba-lab/disteval
Contact map prediction	QDeep [68]	ResNets	Distance-based single-model protein quality estimation method based on residue-level ensemble error classifications.	https://github.com/Bhattacharya-Lab/QDeep
	ResPRE [44]	Deep residual convolutional neural network	ResPRE is better than the methods that are built on co-evolution coupling analyses or a meta-server based neural network	https://zhanglab.ccmb.med.umich.edu/ResPRE [W], https://github.com/leeyang/ResPRE.
	MapPred [45]	Deep ResNet	Covariance features derived from MSA are used to predict contact maps, distance maps, and distance distribution	http://yanglab.nankai.edu.cn/mappred/ [W]
	DEEPCON [47]	ResNet, U-Net, and FCN	Compares various deep learning architectures for protein contact prediction	https://github.com/badriadhikari/DEEPCON/
	DeepECA [48]	CNN with ResNet	Structures predicted by DeepECA, based on contacts and SS, are more accurate than existing evolutionary coupling analysis methods	https://github.com/tomiilab/ DeepECA
	ContactGAN [50]	GAN	GAN-based denoising framework to push the limit of protein contact prediction	https://github.com/largelymfs/deepcontact
	InterPretContactMap [70]	Attention based CNN	Attention mechanisms was used to improve the interpretability of deep learning contact prediction models.	https://github.com/jianlin-cheng/InterpretContactMap
	TripletRes [54]	ResNet	TripletRes model inputs are raw co-evolutionary features, and it predicts high-accuracy contact maps	https://zhanglab.ccmb.med.umich.edu/TripletRes/ [W]
Overall protein structure prediction pipeline	AlphaFold [12]	Deep Neural Network	Accurate predictions of the distances between pairs of residues, which convey more information about the structure than contact predictions	https://github.com/deepmind/deepmind-research/tree/master/alphafold_casp13
	trRosetta [56]	ResNet	A fully Tensorflow-based open-source implementation to predict distograms; demonstrated to outperform AlphaFold	https://github.com/gjoni/trRosetta https://yanglab.nankai.edu.cn/trRosetta/ [W]
	RaptorX [11,43]	ResNet	The original RaptorX method upgraded to predict distograms	http://raptorx.uchicago.edu/AbInitioFolding/ [W], https://github.com/j3xugit/RaptorX-Contact
	MULTICOM [75]	Deep Convolutional neural network	Predicts protein structure, secondary structure, solvent accessibility, disorder region, as well as contact map	http://sysbio.rnet.missouri.edu/multicom_cluster/ [W]
	C-I-TASSER and C-QUARK [64]	Deep residual CNN	C-I-TASSER is derived from I-TASSER for high-accuracy protein structure and function predictions.	https://zhanglab.ccmb.med.umich.edu/C-I-TASSER/ [W]
Quality Assessment (QA) and refinements	QDeep [68]	ResNets	QDeep is a new distance-based single-model protein quality estimation method based on residue-level ensemble error classifications.	https://github.com/Bhattacharya-Lab/QDeep
	ResNetQA [69]	ResNet	It is a new single-model-based QA method for both local and global quality assessment.	https://github.com/AndersJing/ResNetQA
	DeepAccNet [72]	3D Convolution, 2D convolutions	DeepAccNet estimates per-residue accuracy and residue–residue distance signed error in protein models and uses these predictions to guide Rosetta protein structure refinement.	https://github.com/hiranumn/DeepAccNet
Single Particle picking or cryo-EM cleaning	PIXER [85]	Deep Neural Network	PIXER is a fully automated particle-selection method, it can acquire accurate results under low-SNR conditions within minutes.	https://github.com/ZhangJingrong/PIXER
	AutoCryoPicker [89]	Unsupervised ML algorithm	AutoCryoPicker can recognize particle-like objects from noisy Cryo-EM micrographs without the need of labeled training data, it is a useful tool for Cryo-EM protein structure determination	https://github.com/jianlin-cheng/AutoCryoPicker
	MicroGraphCleaner [87]	U-net architecture	MicrographCleaner is a tool that automatically discriminates between regions of micrographs which are suitable for particle picking, and those that are not.	https://github.com/rsanchezgarc/micrograph_cleaner_em
	CASSPER [86]	InceptionV4, Residual Network	CASSPER is the first particle picking tool implementing the Residual Network architecture for efficient pixel-wise classification.	https://github.com/airis4d/CASSPER
Structure Prediction in Cryo-EM etc.	Dong Si Method [90]	Cascade CNN	It predicts secondary structure elements, backbone structure, and Cα atoms, combining the results of each to produce a complete prediction map.	https://github.com/DrDongSi/Ca-Backbone-Prediction
	Emap2sec [92]	CNN	Emap2sec identifies the secondary structures of proteins in Electron Microscopy maps at resolutions of between 5 and 10 Å.	https://github.com/kiharalab/Emap2sec
	DeepTracer [93]	Convolutional Network Architecture	DeepTracer determines the all-atom structure of a protein complex based on a Cryo-EM map and amino acid sequence.	https://deeptracer.uw.edu/home
	DEFMap [96]	3D convolution	DEFMap directly extracts the dynamics associated with the atomic fluctuations that are hidden in Cryo-EM density maps.	https://github.com/clinfo/DEFMap
Cryo-EM	EMRefiner [97]	Monte Carlo	It is a Monte Carlo-based method for protein structure refinement and determination using a Cryo-EM density map	https://zhanglab.ccmb.med.umich.edu/EM-Refiner/
	DEMO - EM [98]	Deep Neural Network	DEMO-EM, does structure assembly of multi-domain proteins from Cryo-EM density maps.	https://zhanglab.ccmb.med.umich.edu/DEMO-EM/ [W]
	SuperEM [99]	GAN	SuperEM captures protein structure information from Cryo-EM maps more effectively than raw maps.	https://github.com/kiharalab/SuperEM
Multi Domain Protein Structures	FUpred [102]	ResNet	FUpred has better ability of domain boundary prediction than threading-based and machine learning-based methods.	https://zhanglab.ccmb.med.umich.edu/FUpred/ [W]

W: web server.