Accurate prediction of drug absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties is crucial for reducing clinical failure rates. In this study, we propose ATLAS-DMPNN (Attention-guided Topological Location-Aware Structural Directed Message Passing Neural Network), an enhanced molecular representation learning framework tailored for ADMET prediction. This framework systematically addresses three major limitations of directed message passing neural networks (D-MPNN): uneven feature weight allocation, insufficient structural sensitivity, and limited capability in identifying critical substructures. The framework integrates three innovations: an attention-guided message passing mechanism, topological location-aware atomic representations, and ant colony optimization-based structural motif mining. Experimental results demonstrate that ATLAS-DMPNN significantly outperforms the standard D-MPNN in classification tasks, achieving an average performance improvement of approximately 5.1%, especially in complex toxicity and metabolism prediction tasks. This work not only enhances the accuracy and intcrprctability of ADMET predictions but also introduces a novel molecular representation learning paradigm for computer-aided drug design, promising to accelerate drug discovery and reduce development costs. © 2025 IEEE.

Deploying recommender systems in edge computing environments poses significant challenges due to constrained computational resources. To address this issue, this paper proposes an Adaptive Elite Evolutionary Discrete Particle Swarm Optimization algorithm (AEE-DPSO) designed for efficient edge-side recommendation optimization. The method incorporates multiple adaptive mechanisms, including perturbation level control based on differential roulette wheel selection, dynamic guidance adjustment between personal and global optima, and exponentially decaying perturbation intensity. These mechanisms effectively reduce the search space dimension during optimization, mitigating the curse of dimensionality and enhancing convergence efficiency. Furthermore, the algorithm integrates fine-grained micro-perturbation strategies and difference-based dimension filtering to refine local exploitation. Experimental results show that, compared to the current high-performance baseline algorithm (EEDPSO), the proposed AEE-DPSO achieves either improved or comparable fitness while significantly reducing time complexity, thereby improving computational efficiency on edge devices. ©2025 IEEE.

The covert communication has emerged as a critical technique to prevent eavesdropping and illegal detection in wireless networks. In this paper, we propose a novel covert communication scheme based on index modulation (IM) for Space-Time Block Code (STBC)-assisted simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) system. First, STAR-RIS system based on STBC is used to transmit covert information, and the private bits are decomposed into index bits and information bits, which are transmitted through index modulation. Second, in the process of modulating the information bits, the artificial noise (AN) generated based on the channel information of legitimate receivers, it can not only be used to conceal the modulation symbols of the information bits, but also interfere to illegal eavesdroppers. Finally, the simulation results verify the effectiveness and security of the proposed scheme. ©2025 IEEE.