Computational models

Computational models are generated using various computational algorithms and rely on existing biological data, primarily the amino acid sequence of a protein or the nucleotide sequence of nucleic acids. They do not rely on experimental scattering or resonance data from the molecular being modelled, although they often learn from existing experimental structures.

Homology modelling

This method, also known as comparative modelling, is based on the principle that if two proteins have sufficiently similar sequences, they likely fold into similar 3D structures. Homology modelling predicts the structure of a “target” protein using the known experimental structure of a related protein (i.e. template) as a guide. The accuracy of the resulting model is highly dependent on the degree of sequence similarity between the target and template (higher similarity generally means higher expected accuracy) and the quality of the template structure itself.

De novo or ab initio

This approach aims to predict a protein’s 3D structure solely from its amino acid sequence, without relying on a pre-existing structural template of a related protein. Historically, this was challenging, especially for larger proteins.

However, recent significant breakthroughs, particularly with methods based on deep learning like AlphaFold and RosettaFold, have dramatically improved the accuracy of de novo prediction for many proteins, often achieving quality comparable to experimental structures. Nevertheless, the accuracy of these predictions can still vary depending on the protein, and predicting the structures of complexes or highly flexible regions remains an active area of method development.

For more information about AlphaFold, you can explore the dedicated online training module available, AlphaFold: A practical guide.