PUResNet: prediction of protein-ligand binding sites using deep residual neural network

Abstract

Background: Predicting protein-ligand binding sites is a fundamental step in understanding the functional characteristics of proteins, which plays a vital role in elucidating different biological functions and is a crucial step in drug discovery. A protein exhibits its true nature after binding to its interacting molecule known as a ligand that binds only in the favorable binding site of the protein structure. Different computational methods exploiting the features of proteins have been developed to identify the binding sites in the protein structure, but none seems to provide promising results, and therefore, further investigation is required.

Results: In this study, we present a deep learning model PUResNet and a novel data cleaning process based on structural similarity for predicting protein-ligand binding sites. From the whole scPDB (an annotated database of druggable binding sites extracted from the Protein DataBank) database, 5020 protein structures were selected to address this problem, which were used to train PUResNet. With this, we achieved better and justifiable performance than the existing methods while evaluating two independent sets using distance, volume and proportion metrics.

Keywords: Binding site prediction; Convolutional neural network; Data cleaning; Deep residual network; Ligand binding sites.

Fig. 1

Flow diagram of data cleaning process

Fig. 2

Flow diagram showing calculation of Tanimoto index

Fig. 3

Model PUResNet architecture showing both encoder and decoder block with skip connections

Fig. 4

Success rate plot for different DCC values combining all fold (kalasanty vs PUResNet)

Fig. 5

Histogram of DVO values combining all folds (kalasanty vs PUResNet)

Fig. 6

Success rate plot for different DCC values in Coach420 dataset (PUResNet vs kalasanty)

Fig. 7

Histogram of DVO values for protein structure having DCC $\le$ 4 Å in Coach420

Fig. 8

Histogram of PLI values for protein structure having DCC $\le$ 4 Å in Coach420

Fig. 9

Success rate plot for different DCC values in BU48 dataset (PUResNet vs kalasanty)

Fig. 10

Histogram of DVO values for protein structure having DCC $\le$ 4 Å in BU48

Fig. 11

Histogram of PLI values for protein structure having DCC $\le$ 4 Å in BU48

Fig. 12

Scatter plot showing DCC values of Coach420 dataset predicted by kalasanty and PUResNet with different views(I-V), View I showing DCC values $\le$ 20 Å from PUResNet and kalasanty, View II showing DCC values $\le$ 10 Å from PUResNet and kalasanty, View III showing DCC values $\ge$ 120 Å from PUResNet and $\le$ 20 Å from kalasanty, View IV showing DCC values $\ge$ 120 Å from kalasanty and $\le$ 20Å from PUResNet and View V showing DCC values $\ge$ 124 Å from kalasanty and PUResNet

Fig. 13

Scatter plot showing DCC values of BU48 dataset predicted by kalasanty and PUResNet with different views (I–V), View I showing DCC values $\le$ 20 Å from PUResNet and kalasanty, View II showing DCC values $\le$ 10 Å from PUResNet and kalasanty, View III showing DCC values $\ge$ 120 Å from PUResNet and $\le$ 20 Å from kalasanty, View IV showing DCC values $\ge$ 120 Å from kalasanty and $\le$ 20 Å from PUResNet and View V showing DCC values $\ge$ 120 Å from kalasanty and PUResNet

Fig. 14

Protein strucutre ( 2zhz, 3h39, 3gpl, 7est, 2w1a, 1a4k) from Coach420, showing predicted binding site by kalasanty(Blue region) and PUResNet (Red Region)

Fig. 15

Bound and Unbound pair ((1a6u,1a6w), (1gcg,1gca)), showing predicted binding site by kalasanty(Blue region) and PUResNet (Red Region)