Method for identifying binding areas of protein complexes

Patent number: 2743316

Authors: Peter Popov; Igor Kozlovsky

Patent owner: Skolkovo Institute of Science and Technology

Proteins found in a living organism are an important «application point» for the action of drugs since proteins are responsible for many functions of an organism. The interaction of a drug and a protein always takes place in three-dimensional space, where the spatial structure of the protein determines whether a drug can change the structure of the protein in such a way that it performs its biological function. Therefore, if we know the exact three-dimensional structure of the protein, we can predict the probability of binding of proteins with the drug, which means we can evaluate the effectiveness of treatment or prevention.

The technology developed by Russian researchers is based on machine learning technology. The algorithm is based on a digital analysis of the structures of protein complexes, allowing to estimate the density of atoms that make up the complexes. On the basis of this analysis, a voxel grid is constructed. A voxel grid is a three-dimensional model of a complex divided into smaller cubic grids; each cell of a voxel grid has a predictive value of possible binding, called confidence level. Once we have the results of the calculated prognosis (more precisely, the prediction of the centers of the binding areas), it is possible to make a prognosis of treatment or prevention, taking into account amino acid residues. The advantage of this «prediction system» is the constant accumulation and updating of data on received forecasts. In addition, it allows individual analysis of specific proteins of each patient, which means selection of the most effective personalized treatment. For example, the choice of the dosage of the drug or the duration of its use is much more effective to carry out not on the basis of averaged treatment protocols, but taking into account individual characteristics of the patient’s body.