Machine learning study of the extended drug-target interaction network informed by pain related voltage-gated sodium channels

Abstract

Pain is a significant global health issue, and the current treatment options for pain management have limitations in terms of effectiveness, side effects, and potential for addiction. There is a pressing need for improved pain treatments and the development of new drugs. Voltage-gated sodium channels, particularly Nav1.3, Nav1.7, Nav1.8, and Nav1.9, play a crucial role in neuronal excitability and are predominantly expressed in the peripheral nervous system. Targeting these channels may provide a means to treat pain while minimizing central and cardiac adverse effects. In this study, we construct protein-protein interaction (PPI) networks based on pain-related sodium channels and develop a corresponding drug-target interaction network to identify potential lead compounds for pain management. To ensure reliable machine learning predictions, we carefully select 111 inhibitor data sets from a pool of more than 1000 targets in the PPI network. We employ 3 distinct machine learning algorithms combined with advanced natural language processing (NLP)-based embeddings, specifically pretrained transformer and autoencoder representations. Through a systematic screening process, we evaluate the side effects and repurposing potential of more than 150,000 drug candidates targeting Nav1.7 and Nav1.8 sodium channels. In addition, we assess the ADMET (absorption, distribution, metabolism, excretion, and toxicity) properties of these candidates to identify leads with near-optimal characteristics. Our strategy provides an innovative platform for the pharmacological development of pain treatments, offering the potential for improved efficacy and reduced side effects.

Figure 1.

The flowchart of screening nearly optimal lead compounds for inhibiting pain-related voltage-gated sodium channels (VGSCs). (A) Protein–protein interaction (PPI) networks of the 4 VGSCs involve more than 1000 proteins, including 4 treatment targets SCN3A, SCN9A, SCN10A, and SCN11A. Each of them has a core and global PPI network. Further details of the PPI networks are provided in the Table S1 in the Supporting Information (available at: http://links.lww.com/PAIN/B940). (B) The drug–target interaction (DTI) network involves 111 targets and 150,147 inhibitor compounds. Here, only 4 treatment targets (SCN3A, SCN9A, SCN10A, and SCN11A) with several compounds are displayed for simplicity. The yellow dashed lines indicate the connections among 111 targets. (C) Predictive models for side effect and repurposing evaluation, as well as ADMET screening. ADMET, absorption, distribution, metabolism, excretion, and toxicity.

Figure 2.

The heatmap of cross-target binding affinities (BAs) predictions for the extended DTI networked informed by 4 pain-related voltage-gated sodium channels. The left labels of the heatmap represent all the inhibitor data sets and those above the heatmap mean the machine learning (ML) models. The diagonal elements in the heatmap denote the Pearson correlation efficient (R) of 10-fold cross-validation for all the ML models. The off-diagonal elements in each row indicate the highest BA values of inhibitors of one data sets predicted by 111 ML models. This heatmap is used to reveal the inhibitor specificity of each data set on other protein targets. DTI, drug–target interaction.

Figure 3.

Examples of predictions of side effects and repurposing potentials. (A) The first row, second row, and third row represent example inhibitor data sets of 2 treatment targets SCN9A and SCN10A that have side effects on none, 1, and 2 of the given 2 side effect targets, respectively. The blue frames indicate where there are no side effects. (B) Displays example inhibitor data sets of side effect targets that are equipped with repurposing potentials on treatment targets SCN9A and SCN10A. The yellow frames indicate that the inhibitors have repurposing potential for one treatment target but have no side effect on the other treatment target.

Figure 4.

Three examples of correlated predicted BA values suggesting the structure and/or sequence similarities of proteins. In each panel, the x-axis and y-axis represent the predicted BA values on 2 other proteins, and the scattered points with colors indicate the experimental labels of inhibitors of the target. The 3D structure alignment is shown in the right of the panel, and the 2D sequence alignment is shown below. In the 3D structure alignment, PDB 6ZG4 and 3UON are used for CHRM1 and CRMH2, PDB 6QY7 and 6QY9 for CSNK2A1 and CSNK2A2, PDB 3ELJ, 7N8T, and 3KVX for MAPK8, MAPK9, and MAPK10, respectively. BA, binding affinities.

Figure 5.

Druggable property screening based on ADMET properties, synthesizability, and hERG side effects on compounds from 5 protein data sets: SCN5A, SCN9A, SCN10A, CNR1, and SRC. The colors of the points indicate the experimental BAs for these targets. The x- and y-axis represent various predicted ADMET properties, synthesizability, or hERG side effects. Blue frames highlight the optimal ranges of these properties and side effects. ADMET, absorption, distribution, metabolism, excretion, and toxicity.

Figure 6.

Assessment of 13 ADMET properties for those molecular compounds with repurposing potentials. (A and B) indicate the evaluations of ADMET properties of 2 compounds CHEMBL1767278 and CHEMBL1453498, and C and D represent their chemical graphs and predictions of side effects, respectively. The boundaries of yellow and red regimes in A and B show the upper and lower limits of the optimal ranges for 13 ADMET properties, respectively. The blue curves suggest values of the specified 13 ADMET properties. The details of these property abbreviations are as following: MW, molecular weight; logP, log of octanol/water partition coefficient; logS, log of the aqueous solubility; logD, logP at physiological pH 7.4; nHA, number of hydrogen bond acceptors; nHD, number of hydrogen bond donors; TPSA, topological polar surface area; nRot, number of rotatable bonds; nRing, number of rings; MaxRing, number of atoms in the biggest ring; nHet, number of heteroatoms; fChar, formal charge; nRig, number of rigid bonds; ADMET, absorption, distribution, metabolism, excretion, and toxicity.

Figure 7.

The docking structure of our 2 optimal lead compounds bound to 2 pain targets SCN0A and SCN10A, and their 2D interaction diagrams. We use AutoDock Vina to implement the protein–ligand docking and find the hydrogen bonds generated during the docking of 2 compounds.