Russian scientists from Skoltech created a batenetpp convolutional neural network, which by the protein structure predicts the location of the binding sites with peptides, which are intermediaries and regulators of 40 percent of cellular processes. A new approach, which is told in a press release on the SDUNUKI website.rf, will allow to look for peptides suitable for creating new drugs.
Bitenetpp was trained in a set of data on proteins and small molecules, and then returned on the set of proteins and peptides. The detection of binding sites is similar to the search for an object in the picture, however, a non-flat image is applied to the input, and the three-dimensional atomic structure of the protein, and the model analyzes not pixels, but their three-dimensional analogs – voxels.
Neuraleta is capable not only with high accuracy to detect the binding sites of proteins, but also seeking them strikingly quickly: for the analysis of one protein structure it takes less than a second.
Peptide molecules are an alternative to large medicines that may not get to the desired protein or are not capable of providing the impact necessary for it. Peptides, on the contrary, have high specificity and affinity, that is, they have the ability to effectively bind to molecules, changing how they interact with other proteins. The effect on these interactions can be a way to treat a variety of diseases.