Link prediction on graphs has applications spanning from recommender systems to drug discovery. Temporal link prediction (TLP) refers to predicting future links in a temporally evolving graph and adds additional complexity related to the dynamic nature of graphs. State-of-the-art TLP models incorporate memory modules alongside graph neural networks to learn both the temporal mechanisms of incoming nodes and the evolving graph topology. However, memory modules only store information about nodes seen at train time, and hence, such models cannot be directly transferred to entirely new graphs at test time and deployment. In this work, we study a new transfer learning task for temporal link prediction and develop transfer-effective methods for memory-laden models. Specifically, motivated by recent research showing the informativeness of structural signals for the TLP task, we augment a structural mapping module to the existing TLP model architectures, which learns a mapping from graph structural (topological) features to memory embeddings. Our work paves the way for a memory-free foundation model for TLP.

MOTIVATION: Unraveling the human interactome to uncover disease-specific patterns and discover drug targets hinges on accurate protein-protein interaction (PPI) predictions. However, challenges persist in machine learning (ML) models due to a scarcity of quality hard negative samples, shortcut learning, and limited generalizability to novel proteins. RESULTS: In this study, we introduce a novel approach for strategic sampling of protein-protein noninteractions (PPNIs) by leveraging higher-order network characteristics that capture the inherent complementarity-driven mechanisms of PPIs. Next, we introduce Unsupervised Pre-training of Node Attributes tuned for PPI (UPNA-PPI), a high throughput sequence-to-function ML pipeline, integrating unsupervised pre-training in protein representation learning with Topological PPNI (TPPNI) samples, capable of efficiently screening billions of interactions. By using our TPPNI in training the UPNA-PPI model, we improve PPI prediction generalizability and interpretability, particularly in identifying potential binding sites locations on amino acid sequences, strengthening the prioritization of screening assays and facilitating the transferability of ML predictions across protein families and homodimers. UPNA-PPI establishes the foundation for a fundamental negative sampling methodology in graph machine learning by integrating insights from network topology. AVAILABILITY AND IMPLEMENTATION: Code and UPNA-PPI predictions are freely available at https://github.com/alxndgb/UPNA-PPI.

Software Defined Storage (SDS) separates the control layer from the data layer allowing the automation of data management and deployment of Commercial Off-The-Shelf (COTS) storage media rather than expensive traditional hardware-based solutions. Cloud block storage services lack an SDS framework that allows customization of block storage, policy enforcement, automate provisioning, and storage management. SDS decreases the human intervention and improves the resource utilization. Moreover, SDS allows cloud tenants to define customized functionalities based on their needs with guaranteed performance and high availability that meets Service Level Agreements (SLAs). However, maintaining SLAs requirements in cloud block storage is challenging due to the storage cluster features, the workload interference, the workload characteristics and other indirect related latent variables. To address the mentioned issues, cloud providers often over-provision the storage resources. Moving towards SDS, we initiate a framework for cloud block storage as an active storage system. Our framework provides customization of block storage services and optimized scheduling decisions based on the workload characteristics and performance of the underlying data layer leveraging a self-learning scheduler. The proposed scheduler treats the storage backend nodes as a black box and requires zero knowledge of their internal states. We showcase a practical application of the proposed scheduler in our private OpenStack deployment.

A major requirement of cloud block storage services is guaranteed performance and high availability. However, offering guaranteed Service Level Agreements (SLAs) in cloud block storage services is often not straightforward. Cloud block storage performance may be affected by physical disk background operations, like garbage collection, storage cluster features, workload interference and the chraracteristcs of the workload itself. On the other hand, the underlying physical storage drives do not expose the internal states to higher level block storage service offerings. Therefore, SLAs can only be satisfied by over-provisioning the storage resources. To address this issue, we propose a self-learning scheduler that can dynamically adapt based on the workload, and efficiently provide a scheduling decision with zero knowledge of the underlying hardware. We study two candidate algorithms based on Feedback learning and Two-Phase learning. We used workloads that were deducted from real-world block-level traces of an enterprise data center, and conducted extensive simulations. Our results indicate that the self-learning scheduling approach can reduce the SLA violations by mitigating the unexpected resource fluctuation, and the scheduler can also adapt dynamically to various workloads.