Investigating a Closed-Loop Agentic AI Framework for Intelligent Decision-Making in High-Mix Electronic Manufacturing Services (EMS) Supply Chain Transformation

268,065.00

The exploration of drug–target interactions (DTI) is a critical stage in the drug development pipeline, as it enables researchers to understand how chemical compounds interact with biological targets and influence therapeutic outcomes. Traditionally, this process is time-consuming and costly, often requiring extensive laboratory experiments. With the advancement of computational techniques, particularly deep learning, researchers have been able to significantly accelerate DTI prediction and reduce experimental burden. Widely used deep learning models such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs) have demonstrated promising results in extracting features from sequential and spatial data. However, these approaches are limited in their ability to effectively capture the inherent graph-based structure of molecular data, where atoms and bonds naturally form complex relational networks.

To address this limitation, graph neural networks (GNNs) have emerged as a powerful alternative, as they are specifically designed to operate on graph-structured data. GNNs enable more meaningful representation learning by directly modeling the chemical and structural properties of molecules. In this work, we propose Deep Neural Computation (DeepNC), an advanced deep learning framework for DTI prediction that integrates multiple GNN architectures, including Generalized Aggregation Networks (GENConv), Graph Convolutional Networks (GCNConv), and Hypergraph Convolution with Hypergraph Attention (HypergraphConv). These models collaboratively enhance the ability to capture both local and global structural dependencies within molecular graphs.

In the DeepNC framework, drug compounds are encoded through stacked GNN layers, while protein targets are processed using one-dimensional convolutional neural networks to extract sequence-based features. The learned representations of drugs and targets are then concatenated and passed through fully connected layers to predict binding affinity values. Experimental evaluation on benchmark datasets such as Davis and KIBA, along with an independently constructed Allergy dataset, demonstrates that DeepNC consistently outperforms baseline methods. Specifically, it achieves improved performance in terms of mean squared error (MSE) and concordance index (CI), highlighting its effectiveness and robustness. Overall, DeepNC provides a scalable and accurate approach for advancing computational drug discovery and precision medicine