A novel protein descriptor for the prediction of drug binding sites

Abstract

Background: Binding sites are the pockets of proteins that can bind drugs; the discovery of these pockets is a critical step in drug design. With the help of computers, protein pockets prediction can save manpower and financial resources.

Results: In this paper, a novel protein descriptor for the prediction of binding sites is proposed. Information on non-bonded interactions in the three-dimensional structure of a protein is captured by a combination of geometry-based and energy-based methods. Moreover, due to the rapid development of deep learning, all binding features are extracted to generate three-dimensional grids that are fed into a convolution neural network. Two datasets were introduced into the experiment. The sc-PDB dataset was used for descriptor extraction and binding site prediction, and the PDBbind dataset was used only for testing and verification of the generalization of the method. The comparison with previous methods shows that the proposed descriptor is effective in predicting the binding sites.

Conclusions: A new protein descriptor is proposed for the prediction of the drug binding sites of proteins. This method combines the three-dimensional structure of a protein and non-bonded interactions with small molecules to involve important factors influencing the formation of binding site. Analysis of the experiments indicates that the descriptor is robust for site prediction.

Keywords: Binding sites prediction; Deep learning; Molecule descriptor; Protein pockets.

Fig. 1

Performance comparison of different channels

Fig. 2

Comparison of different Minpts values for DBSCAN

Fig. 3

5-fold crossover experiment for Top3 prediction. a fold1 b fold2 c fold3 d fold4 e fold5

Fig. 4

5-fold crossover experiment for Top5 prediction. a fold1 b fold2 c fold3 d fold4 e fold5

Fig. 5

Error sum of different methods for Top3 predictions

Fig. 6

Error sum of different methods for Top5 predictions

Fig. 7

Generalization effect on PDBbind of the model trained using scpdb dataset

Fig. 8

A slightly modified version of LIGSITE. The voxels represent the solvent, the green dots are the protein atoms, and the white area is the protein contour. The red lines are the scanning lines in the x direction with a step of 1Å. When a scanning line experiences a protein-solvent-protein event, the voxel contained in the intermediate solvent undergoes a PSP event indicated by the purple voxels. In three-dimensional case, proteins are scanned in seven directions including x,y,z and four diagonal directions

Fig. 9

The calculation process of the van der Waals force channel grid. The probe is placed in each grid voxel in turn, and the van der Waals potential between the probe and the protein is calculated as the voxel value

Fig. 10

Determination of the positive samples. The black dot is the geometric center of the protein binding site, and a square (red block in the figure) with a side length of 20Å centered on it is set as the positive sample area; the total may include 4×4×4=64 sampling blocks, which are marked as positive samples

Fig. 11

Training flow chart (4 channels)