TABLE 1.

Recent computational tools considered approachable to non‐specialist protein engineers.

Tool	Functionality	Features and limitations	Reference
AlphaFold 2.0	Computational tool designed for predicting three‐dimensional protein structure using a deep learning system trained on structures in the PDB.	Features: Provides protein structure prediction with high accuracy. Limitations: Does not model the dynamics and mechanics of the protein. Low confidence in unstructured regions.	[12]
ROSIE	The gold‐standard, a web platform hosting dozens of tools from the Rosetta suite of programs that enable modelling and protein design, including: Molecular docking, prediction and design for protein stability and solubility.	Features: Enables access to multiple tools a protein engineer would require from the Rosetta suite under one user‐friendly environment. Limitations: Platform has been developed by different groups over time, thus it can be difficult to implement and utilise different modules seamlessly. Deep investigation into limitations provided by [13].	[14]
HotSpot Wizard 3.0	Web application that allows identification of hot spots for mutagenesis.	Features: User can input either the sequence or three‐dimensional structure of the target protein. The tool enables development of smart libraries, integrating considerations on function, stability and evolutional variability. Limitations: The aim of the tool is to identify highly mutable functional residues that are unlikely to impair function, thus the design philosophy may not be in line with all protein engineering strategies. For example, the library generated may not be effective if screening for altered ligand specificity.	[15]
FuncLib	To redesign an active site and create multiple‐point designs. Based on conservation analysis and energy calculations.	Features: Algorithm designed specifically to output a small ranked set of stable, multipoint active‐site mutants that are functionally diverse, enabling efficient low‐throughput wet‐lab testing. Limitations: As input, it requires a molecular structure and diverse set of sequence homologues. It works better with a pre‐stablised protein scaffold. In the absence of sufficient knowledge of the protein being engineered, poor results are likely.	[16]
CaverWeb 1.2	To calculate trajectory and interaction energy profiles of a ligand travelling through a protein tunnel.	Features: Integration with other CaverSuite tools, including Caver (software for identification and geometric analysis of tunnels) and CaverDock (software for docking‐based analysis of ligand transport) allows deeper analysis of the ligand transport process by users with limited bioinformatics knowledge and experience using computational tools. Found to be more robust and provide higher resolution than other state‐of‐the‐art tools such as SLITHER and MoMA‐LigPath (doi: 10.1093/bioinformatics/btz386). Effective without extensive knowledge of studied system. Limitations: Lacks the possibility of calculating pores.	[17]
DockingApp (Autodock Vina)	Platform‐independent application for setting up, performing and analysing results from AutoDock Vina.	Features: Provides a user‐friendly graphical interface, enabling easier access to AutoDock Vina. Limitations: Slow docking procedure compared to EquiBind.	[18]
LoopGrafter	For transplanting loops between two structurally related proteins, with a focus on the analysis of dynamic properties of the selected loops to transplant.	Features: Enables optimised transplant of loops between structurally related but functionally different proteins. Limitations: Necessitates both the template and the scaffold proteins as inputs to function.	[19]
Protein Repair One‐Stop‐Shop (PROSS)	Automated web platform aimed at improving protein thermostability and functional yield.	Features: Was found to be successfully implemented by scientists without a background in protein designs. Method has been found reliable enough to only require screening of a limited number of output designs [20]. Limitations: Structure is required for stability calculations, although this could be computationally generated if not available this may not always be reliable.	[21]
SoluProt	Predicts the solubility of a protein specified by input sequence, in Escherichia coli.	Features: Only sequence required as input. Has a user‐friendly web‐server. Has the potential to reduce the cost of experimental studies via proritisation of highly soluble proteins. Limitations: Although higher than predecessors, accuracy is 58.5%.	[22]
DeepSoluE	Predicts the solubility of a protein specified by input sequence, in Escherichia coli.	Features: Similar advantages to SoluProt, with improved accuracy, 59.5%. Limitations: Focus on only E.coli	[23]