Info box 5: Molecular modeling

The first step in modeling the structure of a protein is de novo (ab initio) modeling of the structure from the amino acid sequence without prior knowledge about the spatial arrangement of the amino acids [77]. This approach predicts a protein’s folding based on physical/chemical principles without making use of explicit homolog or template structures in contrast to template-based algorithms. Some successful de novo approaches according to the thirteen Critical Assessment of Techniques for Protein Structure Prediction (CASP13) include MULTICOM [78], SWISS-MODEL [79], QUARK [80], and Rosetta [81].

Predicting the structure of a protein can also be addressed by comparative modeling approaches when there is a suitable template or homologous structure that can be used to guide the process. Comparative modeling approaches mainly align the sequence of two (or more) proteins and use the template structure(s) for the similar parts and try to model the gaps by de novo modeling or other fragment-based approaches. In this category, some of the popular softwares are RosettaCM [82], Modeler [83], HHpred [84], and I-TASSER [85].

In addition to the aforementioned methods, models or low-resolution experimental data from NMR or X-ray crystallography can be improved and refined by several computational techniques such as loop-modeling approaches like next-generation KIC (NGK) [86], and DaReUS-Loop [87, 88] or experimental data-based protocols of Rosetta such as RosettaES [89], CS-Rosetta [90], and RosettaNMR [91].