Table 2.
Summary of recent computational methods to optimize the protein-based therapeutics toward the clinical stage. (DevAsp: Developability Aspects).
| DevAsp. | Name | Objectives | Methods | Ref. |
|---|---|---|---|---|
| Affinity | DeepAAI | Predicting antibody neutralizability with antigen | Graph convolutional network (GCN) and Convolutional neural network (CNN) | [75] |
| RESP | Identification of high affinity antibodies | Autoencoder | [76] | |
| Machine learning-assisted directed evolution | Machine learning is used to quickly screen a full recombination library in silico by using sequence–fitness relationships randomly sampled from the library | K-nearest neighbors, linear regression, decision trees, random forests, and multilayer perceptrons | [77] | |
| GeoPPI | Predicting the change of binding affinity upon mutations | Self-supervised learning and gradient boosting tree (GBT) | [78] | |
| CeVICA | In vitro VHH domain antibody engineering and nanobody binder selection | CDR-directed clustering analysis | [79] | |
| Building a high-quality model of a protein’s fitness landscape and screening ten million sequences via in silico directed evolution | UniRep (an unsupervised deep learning model) and Lasso-LARS/Ridge/Ridge SR/Ensembled Ridge SR | [80] | ||
| CLADE | Guiding protein engineering and directed evolution | K-means, Louvain, and the ensemble of 17 regression models | [81] | |
| Ens-Grad | Designing CDR of human Immunoglobulin G antibodies with high affinities | Machine learning | [82] | |
| DLAB | Predicting antibody–antigen binding for antigens with no known antibody binders | Structure-based deep learning | [83] | |
| ProAffiMuSeq | Predicting the binding free energy change of protein–protein complexes upon mutation | Regression model | [84] | |
| DeepRank | Classification, e.g., predicting an input PPI as biological or a crystal artifact, and regression, e.g., predicting binding affinities | Convolutional Neural Networks (CNNs) | [85] | |
| mCSM-PPI2 | Predicting effects of missense mutations in protein–protein affinity | Machine learning | [86] | |
| mCSM-AB2 | Predicting the effects of missense mutations on Ab-binding affinity | Machine learning | [87] | |
| mmCSM-AB | Predicting multi-point mutations on antigen binding affinity | Machine learning | [88] | |
| mmCSM-PPI | Predicting changes in PPI binding affinity caused by multiple point mutations | Machine learning | [89] | |
| NetTree | Predicting PPI ΔΔG | Convolutional Neural Networks and gradient-boosting trees | [90] | |
| Ymir | Calculating in silico antibody-antigen affinities | 3D-lattice-based framework | [91] | |
| MutaBind2 | Predict binding affinity changes upon single and multiple mutations | Random forest | [92] | |
| SSIPe | Quantitative estimation of the binding affinity changes (ΔΔGbind) | Protein interface profiles and a physics-based energy function | [93] | |
| EasyE and JayZ | Binding affinity estimation | Guaranteed Cost Function Network algorithms, Rosetta energy functions and Dunbrack’s rotamer library | [94] | |
| PPI-Affinity | Predicting binding affinity | Support Vector Machine | [95] | |
| TopologyNet | Predicting the protein-ligand binding affinities and protein stability changes upon mutation | Element-specific persistent homology (ESPH) method and Convolutional Neural Networks | [96] | |
| \ | Generation of high-affinity antibody sequences from low-N training data | Machine learning-based methods | [97] | |
| Predicting binder and non-binder antibodies | Convolutional Neural Networks | [98] | ||
| In silico affinity maturation | Homology Modelling and Protein Docking | [99] | ||
| Selectivity | Computational counterselection | Identifying nonspecific antibody candidates | Multi-task neural network | [100] |
| \ | Determining the polyreactivity status of a given sequence | Support vector machine (SVM) | [101] | |
| \ | Predicting antigen specificity | Deep neural networks | [102] | |
| \ | Assessing polyreactivity from protein sequence | Convolutional Neural Network and Recurrent Neural Network | [103] | |
| \ | Co-optimization of therapeutic antibody affinity and specificity | Deep Learning | [12] | |
| Stability | UniRep | Predicting the stability of natural and de novo designed proteins | Recurrent Neural Network | [104] |
| FoldArchitect | Predicting the stability of proteins | Random forest | [105] | |
| BayeStab | Predicting effects of mutations on protein stability | Graph Neural Network and Bayesian Neural Network | [106] | |
| DynaMut | Predicting the effects of missense mutations on protein stability | Random Forest with Graph-based signatures | [107] | |
| DynaMut2 | Predicting the effects of missense mutations on protein stability | Random Forest with Graph-based signatures | [108] | |
| DeepDDG | Prediction of changes in the stability of proteins due to point mutations | Neural Network | [109] | |
| Clustered tree regression | Predicting the mutation-induced protein folding free energy change | K-means and XGBoost | [110] | |
| PROST | Predicting the ΔΔG upon a single-point missense mutation | XGBoost and Extra-Trees | [111] | |
| PremPS | Evaluating the effects of missense mutations on protein stability | Random forest | [112] | |
| iStable 2.0 | Predicting protein thermal stability changes | Sequence- and structure-based tools | [113] | |
| PON-tstab | Predicting stability of protein variants | Random forests | [114] | |
| ProTstab2 | Predicting Protein Thermal Stabilities | Gradient boosting algorithm | [115] | |
| SimBa | Predicting protein stability changes upon mutations | Multilinear regression | [116] | |
| Scone | Predicting protein stability changes upon mutations | Neural network | [117] | |
| pStab | Predicting protein stability changes upon mutations | Debye–Hückel (DH) formalism | [118] | |
| \ | Predicting the spectra at the unfolding transition and denatured state | Artificial neural network | [119] | |
| \ | Predicting stability of protein variants | MM/GBSA and FEP+ | [120] | |
| \ | Understanding the Stabilizing Effect of Histidine on mAb Aggregation | All-atom molecular dynamics simulations and contact-based free energy calculations | [121] | |
| \ | Predicting protein stability change upon mutations | Variational Auto-Encoders | [122] | |
| \ | Designing mutations located at non-conserved residues with enhanced thermostability | Molecular Dynamics (MD) simulation and energy optimization methods | [123] | |
| \ | Design new sequences translating in functional proteins with enhanced thermal stability | Monte Carlo simulations and Molecular Dynamics (MD) simulation | [52] | |
| Aggregation prevention | Aggregation Time Machine | Assessing the long-term aggregation stability | Monte Carlo Analysis | [124] |
| MAPT | Prediction of antibody aggregation | \ | [125] | |
| ANuPP | Predicting Aggregation Nucleating Regions in peptides and proteins | Logistic regression and Bayesian approach | [126] | |
| MAGRE | Predicting regions prone to protein aggregation | Support Vector Machine | [127] | |
| CORDAX | Detecting APRs and predicts the structural topology and architecture of the fibril core | Logistic regression | [128] | |
| AgMata | Identification of regions involved in aggregation | Machine learning | [129] | |
| \ | Identifying aggregation rate enhancer and mitigator mutations in proteins | Support Vector Machine | [21] | |
| \ | Understanding the relationship between protein aggregation and molecular conformation | Molecular Dynamics (MD) simulation | [130] | |
| Solubility | Aggrescan3D 2.0 | Prediction and engineering of protein solubility | AGGRESCAN | [40] |
| solPredict | Antibody solubility prediction using sequence alone | Machine Learning | [131] | |
| CamSol | Predicting protein solubility | Phenomenological combination of several properties | [132] | |
| ProGAN | Predicting protein solubility | Deep Neural Network and Generative Adversarial Nets | [133] | |
| SKADE | Predicting protein solubility | Neural Network | [134] | |
| PaRSnIP | Predicting protein solubility | Gradient boosting machine | [135] | |
| DeepSol | Predicting protein solubility | Convolutional Neural Network | [136] | |
| SOLart | Predicting protein solubility | Random Forest | [137] | |
| PON-Sol2 | Identifying amino acid substitutions that increase, decrease, or have no effect on the protein solubility | Gradient Boosting algorithm | [138] | |
| SoluProt | Predicting protein solubility | Gradient Boosting algorithm | [139] | |
| SODA | Predicting protein solubility changes upon mutations | Kyte-Doolittle hydrophobicity profile and FESS | [140] | |
| \ | Predicting protein solubility | Machine Learning | [141] | |
| \ | Predicting protein solubility | Machine Learning | [142] | |
| \ | Increasing protein solubility | Variational AutoEncoders | [143] | |
| Viscosity | DeepSCM | Predicting protein viscosity | Convolutional Neural Network | [144] |
| \ | Optimization of highly viscous antibodies while maintaining binding affinity and favorable developability profile | Structure-based design | [145] | |
| \ | Predicting protein viscosity | Molecular Modeling and Machine Learning | [146] | |
| \ | Predicting protein viscosity | Multivariate Regression | [147] | |
| \ | Predicting protein viscosity | Molecular Dynamics (MD) simulation and Logistic regression | [13] | |
| \ | Predicting protein viscosity | Coarse-grained models | [148] | |
| \ | Predicting concentration-dependent viscosity curves | Stepwise linear regression | [149] | |
| \ | Selecting antibody candidates with desirable viscosity properties | Coarse-grained simulation | [150] | |
| Deimmunization and humanization | BioPhi | Humanization and humanness evaluation | Sapiens and OASis | [151] |
| Llamanade | Humanization of nanobodies | Large-scale analysis | [152] | |
| BepiPred-2.0 | Predicting B-cell epitopes | Random forest | [153] | |
| Hu-mAb | Suggesting mutations to an input sequence to reduce its immunogenicity | Random forest and LSTM | [154] | |
| iBCE-EL | Identification of B-cell epitopes (BCEs) | Extremely randomized tree (ERT) and Gradient boosting (GB) | [155] | |
| LBCEPred | Predicting B-cell epitopes | Random forest | [156] | |
| EpiSweep | Deimmunizing therapeutic protein | EpiSweep | [157] | |
| MHCEpitopeEnergy | Deimmunization | Monte Carlo-based rotamer packing and sequence design algorithm | [158] | |
| NetMHCpan-4.1 and NetMHCIIpan-4.0 | Predict binding between peptides and MHC-I and MHC-II | Machine learning | [159] | |
| \ | Deimmunization | Multi-objective combinatorial optimization | [160] | |
| \ | Predictions of antibody-specific epitope | Monte Carlo algorithm and machine learning | [161] | |
| \ | Reducing the immunogenicity | Emini surface accessibility, Parker hydrophilicity, and Karplus & Schulz flexibility methods and molecular dynamics simulation | [162] |