Antonio Longa

Metal-organic frameworks (MOFs) are a versatile class of materials with applications in gas storage, separations, and catalysis. Despite extensive research, one of the key factors hindering their broader deployment is the absence of reproducible and scalable synthetic protocols. To address this, we introduce FAIR-MOFs, a database designed to be Findable, Accessible, Interoperable, and Reusable (FAIR), comprising 45{,}700 curated experimental structures, 33{,}361 geometry-optimised structures, and 4{,}161 entries linked to reported synthesis conditions. Analysis of the dataset showed that the propensity of open-metal sites in MOFs is statistically associated with reaction temperature, metal salts, ligands, topology, and metal secondary building unit. Furthermore, we developed a retrosynthetic recommender that captures literature co-usage patterns among solvents, metal salts, and ligands. For any given component, the system suggests compatible reagents and retrieves MOFs prepared under similar conditions. Finally, we trained a graph-based neural network integrated with our MOF deconstruction module to predict most probable metal salts, ligands and solvents directly from 3D structures of experimental or hypothetical MOFs. Using this model, we successfully synthesised MOFs randomly selected from hypothetical MOF databases illustrating the potential of FAIR-MOFs to accelerate the discovery and enable data-driven synthesis of MOFs.

Graph Neural Networks (GNNs) often struggle to capture the link-specific structural patterns crucial for accurate link prediction, as their node-centric message-passing schemes overlook the subgraph structures connecting a pair of nodes. Existing methods to inject such structural context either incur high computational cost or rely on simplistic heuristics (e.g., common neighbor counts) that fail to model multi-hop dependencies. We introduce SP4LP (Shortest Path for Link Prediction), a novel framework that combines GNN-based node encodings with sequence modeling over shortest paths. Specifically, SP4LP first applies a GNN to compute representations for all nodes, then extracts the shortest path between each candidate node pair and processes the resulting sequence of node embeddings using a sequence model. This design enables SP4LP to capture expressive multi-hop relational patterns with computational efficiency. Empirically, SP4LP achieves state-of-the-art performance across link prediction benchmarks. Theoretically, we prove that SP4LP is strictly more expressive than standard message-passing GNNs and several state-of-the-art structural features methods, establishing it as a general and principled approach for link prediction in graphs.

Graph-level regression underpins many real-world applications, yet public benchmarks remain heavily skewed toward molecular graphs and citation networks. This limited diversity hinders progress on models that must generalize across both homogeneous and heterogeneous graph structures. We introduce RelSC, a new graph-regression dataset built from program graphs that combine syntactic and semantic information extracted from source code. Each graph is labelled with the execution-time cost of the corresponding program, providing a continuous target variable that differs markedly from those found in existing benchmarks. RelSC is released in two complementary variants. RelSC-H supplies rich node features under a single (homogeneous) edge type, while RelSC-M preserves the original multi-relational structure, connecting nodes through multiple edge types that encode distinct semantic relationships. Together, these variants let researchers probe how representation choice influences model behaviour. We evaluate a diverse set of graph neural network architectures on both variants of RelSC. The results reveal consistent performance differences between the homogeneous and multi-relational settings, emphasising the importance of structural representation. These findings demonstrate RelSC's value as a challenging and versatile benchmark for advancing graph regression methods.

Following a fast initial breakthrough in graph-based learning, Graph Neural Networks (GNNs) have reached a widespread application in many science and engineering fields, prompting the need for methods to understand their decision process. GNN explainers have started to emerge in recent years, with a multitude of methods both novel or adapted from other domains. To sort out this plethora of alternative approaches, several studies have benchmarked the performance of different explainers in terms of various explainability metrics. However, these earlier works make no attempts at providing insights into why different GNN architectures are more or less explainable or which explainer should be preferred in a given setting. In this survey we fill these gaps by devising a systematic experimental study, which tests 12 explainers on eight representative message-passing architectures trained on six carefully designed graph and node classification datasets. With our results we provide key insights on the choice and applicability of GNN explainers, we isolate key components that make them usable and successful and provide recommendations on how to avoid common interpretation pitfalls. We conclude by highlighting open questions and directions of possible future research.

The analysis of complex and time-evolving interactions, such as those within social dynamics, represents a current challenge in the science of complex systems. Temporal networks stand as a suitable tool for schematizing such systems, encoding all the interactions appearing between pairs of individuals in discrete time. Over the years, network science has developed many measures to analyze and compare temporal networks. Some of them imply a decomposition of the network into small pieces of interactions; i.e., only involving a few nodes for a short time range. Along this line, a possible way to decompose a network is to assume an egocentric perspective; i.e., to consider for each node the time evolution of its neighborhood. This was proposed by Longa et al. by defining the “egocentric temporal neighborhood”, which has proven to be a useful tool for characterizing temporal networks relative to social interactions. However, this definition neglects group interactions (quite common in social domains), as they are always decomposed into pairwise connections. A more general framework that also allows considering larger interactions is represented by higher-order networks. Here, we generalize the description of social interactions to hypergraphs. Consequently, we generalize their decomposition into “hyper egocentric temporal neighborhoods”. This enables the analysis of social interactions, facilitating comparisons between different datasets or nodes within a dataset, while considering the intrinsic complexity presented by higher-order interactions. Even if we limit the order of interactions to the second order (triplets of nodes), our results reveal the importance of a higher-order representation.In fact, our analyses show that second-order structures are responsible for the majority of the variability at all scales: between datasets, amongst nodes, and over time.