Vito Latora

Food webs are central to ecosystem functioning and stability, yet how they respond to climate warming at the substructural level remains unclear. We used a natural temperature gradient (5–25 °C) spanning 14 whole-stream ecosystems in Iceland to test how warming alters food web substructure organisation (core–periphery) and its consequences for ecosystem functioning (energy flux) and stability (robustness). We found no evidence that warming changed the number or proportion of core and peripheral species, but observed internal reorganisation driven by turnover in core species identity and abundance. While warming consistently decreased robustness to secondary extinctions through reduced food web connectance, core reorganisation enhanced both energy flux and robustness to species loss. Dynamic modelling showed ecosystems were disproportionately sensitive to the loss of core taxa, underscoring their structural importance. These results challenge the view of food webs as fixed hierarchies and reveal an adaptive internal architecture under warming, with core dynamics helping buffer ecosystem responses.

Deep learning networks excel at classification, yet identifying minimal architectures that reliably solve a task remains challenging. We present a computational methodology for systematically exploring and analyzing the relationships among convergence, pruning, and quantization. The workflow first performs a structured design sweep across a large set of architectures, then evaluates convergence behavior, pruning sensitivity, and quantization robustness on representative models. Focusing on well-known image classification of increasing complexity, and across Deep Neural Networks, Convolutional Neural Networks, and Vision Transformers, our initial results show that, despite architectural diversity, performance is largely invariant and learning dynamics consistently exhibit three regimes: unstable, learning, and overfitting. We further characterize the minimal learnable parameters required for stable learning, uncover distinct convergence and pruning phases, and quantify the effect of reduced numeric precision on trainable parameters. Aligning with intuition, the results confirm that deeper architectures are more resilient to pruning than shallower ones, with parameter redundancy as high as 60%, and quantization impacts models with fewer learnable parameters more severely and has a larger effect on harder image datasets. These findings provide actionable guidance for selecting compact, stable models under pruning and low-precision constraints in image classification.

Understanding cooperation in social dilemmas requires models that capture the complexity of real-world interactions. While network frameworks have provided valuable insights to model the evolution of cooperation, they are unable to encode group interactions properly. Here, we introduce a general higher‐order networks framework for multi-player games on structured populations. Our model considers multi-dimensional strategies, based on the observation that social behaviours are affected by the size of the group interaction. We investigate the dynamical and structural coupling between different orders of interactions, revealing the crucial role of nested multi-level interactions and showing how such features can enhance cooperation beyond the limit of traditional models with uni-dimensional strategies. Our work identifies the key drivers promoting cooperative behaviour commonly observed in real-world group social dilemmas.

In 2009, many governments implemented austerity measures. This often resulted in restrictions on science funding, increasing competition for funds. Here, we study the allocation of funding across universities by using a competitiveness measure from complexity science based on the bipartite network of universities and scientific subjects. Drawing on a comprehensive data set of 43,430 UK-funded grants between 2006 and 2020, a period covering the onset and relaxation of austerity, we provide two main results. First, we find that the effect of austerity on university grant income depended on research competitiveness: highly competitive universities increased grant income under austerity, while low-ranked universities lost income. Then, we study the resilience of science to austerity exploiting the natural experiment arising from the 2015 UK general election, when austerity was relaxed unexpectedly. We discover that the resilience of researchers at less competitive universities was surprisingly high. The number and size of grant applications grew more after 2015 at less competitive universities, increasing their grant income. Overall, results show that our network-based measure is a better proxy for research competitiveness than aggregate grant income across scientific subjects.

Neurodegenerative diseases are characterized by complex proteins misfolded that propagate within the brain. For instance, current findings highlight the role of 2 specific misfolded proteins in Alzheimer which are believed to spread using brain fibers as highways. Previous studies investigated such spreading by simulation models or machine learning-based predictors which adopt the brain connectome as the underlying spreading network. However, the structural connectome by construction only describes pairwise connections between nodes in a graph. High-order interaction complex networks offer significant advantages over normal graphs because they can capture interactions that go beyond simple pairwise relationships. Protein misfolding and aggregation often involve cooperative behaviors or group dynamics that normal graphs, with their focus on individual edges, cannot adequately represent. The non-linear and multiscale nature of protein misfolding might be better suited to a richer representation of higher-order models. In this study we investigate whether higher-order networks can provide improved fits and explanatory power in this context. More specifically, we employ a simplicial complex contagion model for amyloid beta to predict protein misfolding spread. The simplicial contagion complex produced a mean reconstruction error of 0.030 for Alzheimer’s patients regarding the predicted protein deposition across all brain region in a 2-year horizon and other results, outperforming previous studies, especially for cases in which the misfolded protein were non-increasing steadily. Despite the limited time span, this study highlights the potential of combining advanced network analysis to capture the intricate dynamics of protein aggregation across neural networks. Clinical relevance This study highlights the potential of high-order networks to improve predictions of misfolded protein spread in Alzheimer’s, offering better insight into protein aggregation dynamics.

The microscopic organization of dynamical systems coupled via higher-order interactions plays a pivotal role in understanding their collective behavior. In this paper, we introduce a framework for systematically investigating the impact of the interaction structure on dynamical processes. Specifically, we develop an hyperedge overlap matrix whose elements characterize the two main aspects of the microscopic organization of higher-order interactions: the inter-order hyperedge overlap (non-diagonal matrix elements) and the intra-order hyperedge overlap (encapsulated in the diagonal elements). This way, the first set of terms quantifies the extent of superposition of nodes among hyperedges of different orders, while the second focuses on the number of nodes in common between hyperedges of the same order. Our findings indicate that large values of both types of hyperedge overlap hinder synchronization stability, and that the larger is the order of interactions involved, the more important is their role. Our findings also indicate that the two types of overlap have qualitatively distinct effects on the dynamics of coupled chaotic oscillators. In particular, large values of intra-order hyperedge overlap hamper synchronization by favoring the presence of disconnected sets of hyperedges, while large values of inter-order hyperedge overlap hinder synchronization by increasing the number of shared nodes between groups converging on different trajectories, without necessarily causing disconnected sets of hyperedges.

Recent studies have shown that novel collective behaviors emerge in complex systems due to the presence of higher-order interactions. However, how the collective behavior of a system is influenced by the microscopic organization of its higher-order interactions is not fully understood. In this work, we introduce a way to quantify the overlap among the hyperedges of a higher-order network, and we show that real-world systems exhibit different levels of intra-order hyperedge overlap. We then study two types of dynamical processes on higher-order networks, namely complex contagion and synchronization, finding that intra-order hyperedge overlap plays a universal role in determining the collective behavior in a variety of systems. Our results demonstrate that the presence of higher-order interactions alone does not guarantee abrupt transitions. Rather, explosivity and bistability require a microscopic organization of the structure with a low value of intra-order hyperedge overlap.

Studying how we explore the world in search of novelties is key to understand the mechanisms that can lead to new discoveries. Previous studies analyzed novelties in various exploration processes, defining them as the first appearance of an element. However, novelties can also be generated by combining what is already known. We hence define higher-order novelties as the first time two or more elements appear together, and we introduce higher-order Heaps’ exponents as a way to characterize their pace of discovery. Through extensive analysis of real-world data, we find that processes with the same pace of discovery, as measured by the standard Heaps’ exponent, can instead differ at higher orders. We then propose to model an exploration process as a random walk on a network in which the possible connections between elements evolve in time. The model reproduces the empirical properties of higher-order novelties, revealing how the network we explore changes over time along with the exploration process.

Networks mediate our economic interactions. From spatial networks representing transportation routes, to more abstract networks of similarity among products, we interact, trade, and compete over networks with a complex structure. Nevertheless, classical economic models of how prices are formed under competition, such as the Edgeworth and the Hotelling models, completely neglect the network structure of real markets. Here, we introduce a dynamical model of price formation, where sellers and buyers are placed on the nodes of a network with any arbitrarily complex topology. Buyers are modeled as random walkers with perfect or limited information about the market, and their distribution depends on the prices and positions of sellers. Concurrently, sellers dynamically adjust their prices to maximize their payoff, based on the current distribution of buyers. These two dynamical processes coevolve and we show that, depending on the positions of the sellers and on the level of information available to the buyers, the model can either converge to fixed prices, or produce cycles of different amplitudes and periods, similar to those observed in real-world markets. By unveiling how competition depends dynamically on the structure of a market, our model allows to quantify the strength of competition in real-world markets and also to identify the most profitable locations for a seller. We use urban street patterns of different cities from all over the world, as examples of spatial networks, to illustrate a novel measure of node centrality, which ranks nodes based on the payoff earned by a seller placed on that node.

Construction of minimal metabolic networks (MMNs) contributes both to our understanding of the origins of metabolism and to the efficiency of biotechnological processes by preventing the diversion of flux away from product formation. We have designed MMNs using a novel in silico synthetic biology pipeline that removes genes encoding enzymes and transporters from genome-scale metabolic models. The resulting minimal gene-set still ensures both viability and high growth rates. The composition of these MMNs has defined a new functional class of genes termed Network Efficiency Determinants (NEDs). These genes, whilst not essential, are very rarely eliminated in constructing an MMN, suggesting that it is difficult for metabolism to be re-routed to obviate the need for such genes. Moreover, the removal of NED genes from an MMN significantly reduces its global efficiency. Bioinformatic analyses of the NED genes have revealed that not only do these genes have more genetic interactions than the bulk of metabolic genes but their protein products also show more protein-protein interactions. In yeast, NED genes are predominantly single-copy and are highly conserved across evolutionarily distant organisms. These features confirm the importance of the NED genes to the metabolic network, including why they are so rarely excluded during minimisation.