Understanding protein functions not only helps to deepen our understanding of biological processes, but also provides a crucial biological basis for disease diagnosis, treatment, and drug development. With the development of deep learning, various computational methods have been proposed to improve annotation efficiency. Most existing methods did not consider the inherent flaws of various data, including the inaccuracy of homology data due to protein variations, and the noise of protein-protein interaction data due to incompleteness and diversity of data sources. Moreover, most existing methods combine multimodal data sources by directly concatenating the features extracted from different modalities without considering the collaborative effect of these features. To address these issues, we propose a new deep learning framework, MFPFP, by designing a heterogeneous graph construction strategy that not only better integrates homology and interaction information, but also enriches the graph with potential latent relationships. Based on the constructed graph, we further design a graph attention mechanism that can leverage the collaborative effect of the two data sources for mutual verification, which can automatically extract important features and reduce noise interference. Moreover, we design a tailored gated fusion module to integrate network information and sequence data, ultimately obtaining comprehensive features for prediction. Compared with eight state-of-the-art methods, MFPFP outperforms in both Fmax and AUPR metrics. Ablation experiments further validate the important contribution of each module in the MFPFP model.

Federated learning, as a collaborative model training method that protects data privacy, has become a prominent research focus in machine learning. Currently, many heterogeneous federated learning algorithms have been proposed to address the challenges posed by data and model heterogeneity among clients. However, existing heterogeneous federated learning algorithms typically do not take the specific structural characteristics of models into account. Instead, they tend to rely on simple, generic models for federated learning training, which may lead to insufficient feature extraction capabilities for certain client models, making it difficult to adapt to various data distributions. Moreover, these algorithms often replace the client models with the global model directly, which undermines the generalization ability of the client models and leads to a lack of personalization, preventing the client models from achieving optimal performance. To tackle these issues, we propose a novel heterogeneous federated learning algorithm, FedSA, which combines a structural-aware federated learning optimization algorithm with a client prediction header aggregation algorithm based on precise encoded features. While keeping the parameters of the participating training models unchanged, it enhances the feature extraction capability of the local client and improves the personalization of the client model. Experimental results show that the proposed method achieves significant improvements in accuracy compared to existing methods across multiple datasets and heterogeneous model scenarios.

Semantic Communication (SemCom) has attracted significant attentions due to its potential to enhance communication efficiency and support human-centric services in 6G networks. However, the presence of mismatched background knowledge and dynamic communication channels decreases the performance of SemCom. These issues ultimately lead to a degradation in users’ quality of experience (QoE). To overcome this challenge, a hybrid semantic and bit communication framework is proposed to effectively improve communication performance under mismatched knowledge constraints. Specifically, we design a time division duplex (TDD) SemCom scheme, where the transmitter and the receiver synchronize background knowledge through the uplink transmission to eliminate mismatch constraints. To guide subframe configuration and communication mode selection in the TDD system, a novel QoE model including perceived quality and energy consumption is proposed, and a long-term average QoE maximization problem is further formulated. To solve the proposed NP-hard problem, a joint subframe configuration and communication mode selection algorithm (JSCA) is designed, and the original problem is decomposed into two subproblems. Firstly, the subframe configuration subproblem is transformed into a quasi-concave problem, and the optimal solution is obtained by the bisection method. Secondly, a deep reinforcement learning (DRL)-based approach is designed to select the communication mode for each service. The numerical results validate the effectiveness of JSCA and demonstrate that the proposed hybrid semantic and bit communication scheme can achieve higher QoE compared with fixed schemes, especially in long-term service scenarios.

Task scheduling as an effective strategy can improve application performance on computing resource-limited devices over distributed networks. However, existing evaluation mechanisms for application completion fail to depict the complexity of diverse applications and time-varying networks, which involve dependencies among tasks, computing resource requirements, multi-dimensional quality of service (QoS) constraints, and limited contact duration among devices. Furthermore, traditional QoS-oriented task scheduling strategies struggle to meet the performance requirements without considering differences in satisfaction and acceptance of the application, leading to application failures and resource wastage. To tackle these issues, a quality of experience (QoE) cost model is designed to evaluate application completion, depicting the relationship among application satisfaction, communications, and computing resources over the time-varying distributed networks. Specifically, considering the sensitivity and preference of QoS, we model the different dimensional QoS degradation cost functions for dependent tasks, which are then integrated into the QoE cost model. Based on the QoE model, the dependent task scheduling problem is formulated as the minimization of overall QoE cost, aiming to improve the application performance over the time-varying distributed networks, which is proven Np-hard. Moreover, a heuristic Hierarchical Multi-queue Task Scheduling (HMTS) algorithm is proposed to address the QoE-oriented task scheduling problem among multiple dependent tasks, which utilizes hierarchical multiple queues to determine the optimal task execution order and location according to different dimensional QoS priorities. Finally, extensive experiments demonstrate that the proposed algorithm can significantly improve the satisfaction of applications.

MOTIVATION: Protein sequence database search and multiple sequence alignment generation is a fundamental task in many bioinformatics analyses. As the data volume of sequences continues to grow rapidly, there is an increasing need for efficient and scalable multiple sequence query algorithms for super-large databases without expensive time and computational costs. RESULTS: We introduce Chorus, a novel protein sequence query system that leverages parallel model and heterogeneous computation architecture to enable users to query thousands of protein sequences concurrently against large protein databases on a desktop workstation. Chorus achieves over 100× speedup over BLASTP without sacrificing sensitivity. We demonstrate the utility of Chorus through a case study of analyzing a ∼1.5-TB large-scale metagenomic datasets for novel CRISPR-Cas protein discovery within 30 min. AVAILABILITY AND IMPLEMENTATION: Chorus is open-source and its code repository is available at https://github.com/Bio-Acc/Chorus.

Federated learning (FL) is a prominent paradigm of 6G edge intelligence (EI), which mitigates privacy breaches and high communication pressure caused by conventional centralized model training in the artificial intelligence of things (AIoT). The execution of multimodal federated perception (MFP) services comprises three sub-processes, including sensing-based multimodal data generation, communication-based model transmission, and computing-based model training, ultimately competitive on available underlying multi-domain physical resources such as time, frequency, and computing power. How to reasonably coordinate the multi-domain resources scheduling among sensing, communication, and computing, therefore, is vital to the MFP networks. To address the above issues, this article explores service-oriented resource management with integrated sensing, communication, and computing (ISCC). Specifically, employing the incentive mechanism of the MFP service market, the resources management problem is defined as a social welfare maximization problem, where the concept of “expanding resources” and “reducing costs” is used to enhance learning performance gain and reduce resource costs. Experimental results demonstrate the effectiveness and robustness of the proposed resource scheduling mechanisms.

Implementing either Federated learning (FL) or split learning (SL) over clients with limited computation/communication resources faces challenges on achieving delay-efficient model training. To overcome such challenges, we investigate a novel distributedClustering-basedHybrid fEdEratedSplit lEarning (CHEESE) framework, consolidating distributed resources among clients by device-to-device (D2D) communications, working in an intra-serial inter-parallel manner. InCHEESE, each learning client can form a cluster with its neighboring helping clients via D2D communications to train an FL model collaboratively. Inside each cluster, the model is split into multiple segments via a model splitting and allocation (MSA) strategy, while each cluster member trains one segment. After completing intra-cluster training, a transmission client (TC) is determined from each cluster to upload a complete model to the base station for global model aggregation under allocated bandwidth. Accordingly, an overall training delay cost minimization problem is formulated, involving the following subproblems: client clustering, MSA, TC selection, and bandwidth allocation. Due to its NP-Hardness, the problem is decoupled and solved iteratively. The client clustering problem is first transformed into a distributed clustering game based on potential game theory, where each cluster further investigates the remaining three subproblems to evaluate the utility of each clustering strategy. Specifically, a heuristic algorithm is proposed to solve the MSA problem under a given clustering strategy, while a greedy-based convex optimization approach is introduced to solve the joint TC selection and bandwidth allocation problem. Extensive experiments on practical models and datasets demonstrate thatCHEESEcan significantly reduce training delay costs.