Biological networks

The relationship between genotype and phenotype remains an outstanding question for organism-level traits because these traits are generally complex. The challenge arises from complex traits being determined by a combination of multiple genes (or loci), which leads to an explosion of possible genotype-phenotype mappings. The primary techniques to resolve these mappings are genome/transcriptome-wide association studies, which are limited by the fact that they do not account for how molecular level changes propagate through regulatory networks, and their lack of causal inference and statistical power. Here, we develop an approach that leverages transcriptional data and takes into account the network-structured nature of responses to experimental gene perturbations, and a generative machine learning model to strengthen statistical power. Our implementation of the approach—dubbed TWAVE—includes a variational autoencoder trained on human transcriptional data, which is incorporated into an optimization framework. TWAVE generates trait expression profiles, which we dimensionally reduce by identifying independently varying generalized pathways (eigengenes). We then conduct constrained optimization to combinations of gene perturbations (and thus causal gene sets) whose measured transcriptomic responses best account for trait differences. These optimal combinations enable the construction of co-perturbation networks, where two gene perturbations are connected if the transcriptomic response to the combined perturbations tend to have more effect on phenotype than either of the single gene perturbations. By considering several complex traits, we show that the approach identifies causal genes that cannot be detected by the primary existing techniques. We suggest that the approach be used to design tailored experiments to identify multi-genic targets to address complex diseases.

Huntington’s disease (HD) is a monogenic neurodegenerative disorder caused by an expanded CAG repeat in the HTT gene, leading to cognitive decline, motor dysfunction, and eventual dementia. Although disease etiology is consistent, variability in age of onset and progression is influenced by CAG repeat length and transcriptional regulation. This study explores cis-regulatory variation in HD using a network-based stratification approach to allele-specific expression (ASE) data.We analyzed RNA-seq data from postmortem prefrontal cortex (Brodmann area 9) samples of 20 HD patients (GSE64810). Variant calling and read counting were performed using GATK, followed by ASE analysis. Patients were stratified into clusters based on allelic imbalance profiles using pyNBS, integrating functional gene relationships. Differential ASE and gene expression analysis identified significantly dysregulated genes, further analyzed through Gene Ontology enrichment and network extension using DisGeNET and STRINGdb.Two distinct patient clusters emerged, highlighting transcriptional heterogeneity in HD. Among 71 differentially expressed and imbalanced genes, six (KRAS, CACNA1B, MBP, COX4I1, CAMK2G, DLL1) showed strong connections to HD-related pathways and neurological disorders. Notably, MBP expression, associated with early HD onset in mice, suggests its role as a modifier gene. These findings demonstrate ASE's utility in uncovering transcriptional regulators underlying HD heterogeneity, complementing traditional approaches.

How the extraordinary diversity of microbiomes emerges from the interactions between individual populations is a important unanswered question in microbial ecology. Key to this process are the cross-feeding networks that structure these systems in which populations release metabolic by-products into the environment for other members of the community to use. Understanding how cross-feeding contributes to coexistence presents a major challenge due to the complex structure of cross-feeding networks and the interdependencies they create between populations. In this work we address this problem by using the tools of network science to develop and analyse a structural model of microbial community cross-feeding (Fig. 1A). Our model represents the community as a bipartite- or hypergraph in which consumers and resources are linked by the consumption and secretion of metabolites. We consider feasible community states by introducing simple rules that determine the presence of consumers and resources. Using percolation theory and the generating function formalism we derive expressions for the realised diversity of consumers and metabolites in cross-feeding networks with arbitrary degree distributions. We apply our model to random networks where we find tipping points at which community diversity abruptly changes in response to changes in network structure (Fig 1B \& C). These points correspond to discontinuous percolation transitions and occur due to the collapse and reassembly of the active cross-feeding network. We also consider how robust these cross-feeding networks are to attack, deriving expressions for the state of the community following the removal of consumers and resources from the system.

Protein function and structure are tightly linked. Proteins have a multiscale structure characterised by complex interactions and substructures that unfold across a wide range of spatial and temporal scales. The coupling across scales means that local mutations can lead to global changes in structure and function. Here, we present a computationally efficient framework that uses graph representations of protein structures and integrates multiscale community detection, using Markov Stability (MS), with topological data analysis (TDA), through the Multiscale Clustering Filtration (MCF), to analyse the impact of mutations by measuring the changes they induce in the full multiscale structure of proteins. Our MS+MCF workflow is based on constructing atomistic energy-weighted graphs of proteins from their 3D structures. These graphs are built for the wildtype (WT) as well as for point mutations generated using FoldX. We capture the robustness of partitions at a given scale with the average Normalised Variation of Information (NVI) and the level of hierarchy in the sequence of partitions through the persistent hierarchy. We demonstrate the application of MS+MCF to the closed conformation of adenylate kinase (AdK). We find that mutations known to disrupt protein dynamics and function show significant deviations in NVI compared to the WT, whereas non-disruptive mutations align closely with the WT. Furthermore, the analysis of the persistent hierarchy shows that mutations with significant functional impact induce abrupt jumps at specific characteristic scales. Analysing protein graphs with MS+MCF enables us to detect the structural and dynamical impacts of mutations by measuring the disruption in hierarchical domain organisation and higher-order interactions that govern the multiscale architecture of the protein.

Higher-order networks are attracting large scientific interest in recent years. However, a key challenge is to infer higher-order interactions from data. Triadic interactions are a fundamental type of higher-order interactions that occur when one node regulates the interaction between two other nodes. Triadic interactions are known to be key in ecosystems, in neuronal networks, and in gene regulation networks. However, information theory approaches to mine triadic interactions are still lacking. In this work we explore the fundamental dynamical properties of networks with triadic interactions between continuous variables.We formulate a general model of networks with triadic interactions. Moreover we propose an information theory framework and an algorithm, validated on the triadic model, that we call Triaction, that is able to mine triadic interactions from data.This algorithm is applied on synthetic data as well as Acute Myeloid Leukemia gene expression data finding new candidates for triadic interactions as well as validating already established biological results.In conclusion, our work proposes a new information theoretic approach to mine triadic interactions. In the future, the method could be applied to other types of data including structured missingness.

The amount and quality of sleep individuals get can impact various aspects of human and non-human animal health,ultimately affecting fitness. For wild animals that sleep in groups, individuals may disturb one another, influencing sleepquality and quantity [1, 2], but this aspect of social sleep has been understudied due to methodological challenges. Bycombining longitudinal accelerometer data and tools from network science, we investigate the social dimension of sleepin a wild troop of baboons (Papio ursinus). We test the hypothesis that individual’s social dominance can affect sleepopportunities by explaining sleep patterns as a function of baboons’ hierarchical social structure. First, we show thatthe troop’s night-time sleep (determined by 40Hz acceleration data) is highly synchronized. By modeling sleep as aMarkov chain (Fig 1A), we show that similarly ranked baboons are more synchronized (Fig 1B), and unexpectedly, moredominant baboons experience less and lower-quality sleep. Next, we link night-time sleep dynamics to daytime spatialproximity networks, which we use as proxy for social relationships. We propose that the hierarchy effect is explained byhigher-ranked baboons resting closer to more group members. This is further endorsed through the analysis of a crosscorrelationnetwork (Fig 1C), where dominant individuals exert greater influence on each other’s night-time behaviourcompared to lower-ranked individuals. Our study provides the first empirical evidence for the impact of social hierarchieson sleep in a wild primate, suggesting that dominance status may impose trade-offs between social rank and the qualityand quantity of sleep.

The HuBMAP Human Reference Atlas (HRA) effort [1-3] aims to develop a common coordinate framework (CCF) for the healthy human body, see HRA Portal at https://humanatlas.io. An international team of 1000+ organ experts (anatomists, pathologist, surgeons, biomedical researchers) and ontologists across 20+ consortia are authoring anatomical structures, cell types, and biomarkers (ASCT+B) tables that capture partonomy and typology information of core entities in the human body. The ASCT+B tables are used to revise and extend existing CCF-relevant ontologies. In close collaboration with the National Institute of Allergy and Infectious Diseases (NIAID), a 3D Reference Object Library was compiled (https://3d.nih.gov/collections/hra) that provides semantically annotated 3D representations of major anatomical structures (see samples in Fig. 1) captured in the ASCT+B tables. The HRA can be extended and explored using several interactive user interfaces: The Registration User Interface (RUI, https://apps.humanatlas.io/rui) supports tissue data registration and annotation across 50+ 3D reference organs. The Exploration User Interface (EUI, https://apps.humanatlas.io/eui) supports exploration of semantically and spatially explicit data—from the whole body to the single cell level. The Cell Distance Explorer (CDE, https://apps.humanatlas.io/cde) computes and visualizes distance distributions between different cells, cell types, and anatomical structures and cell types and morphological features. For an introduction to HuBMAP and HRA goals, data, and code visit the Visible Human MOOC (VHMOOC, https://expand.iu.edu/browse/sice/cns/courses/hubmap-visible-human-mooc). This talk details the multi-modal, multi-scale networks of the HRA, how HRA data is published as an open knowledge graph [4] (https://humanatlas.io/api), and how HuBMAP, SenNet, GTEx, and other tissue data is mapped to the HRA to harmonize data at scale in support of precision health and medicine [3].

The seminal works by Denny Borsboom on the network theory of psychopathology, proposing to conceptualize mental disorders as networks of symptoms [1], have been among the most innovative scientific ideas in the mental health field. Building on this theoretical framework, the network theory of plasticity has been recently introduced [2, 3]. This theory proposes the connectivity strength among the elements of a system as a measure of system plasticity and thus of its ability to change its outcome. In particular, the weaker the connectivity, the higher the plasticity of the system. When conceptualizing an individual as a network of interconnected symptoms, the individual’s plasticity -- and thus their ability to transition from psychopathology to wellbeing -- is predicted to be inversely related to the connectivity strength (i.e., the degree of positive or negative co-occurrence) among the symptoms such as the nine standard symptoms for major depressive disorder measured according to DSM-5. The validity of this operationalization has been recently demonstrated. Findings revealed that the baseline connectivity strength among symptoms inversely correlates with subsequent improvement over four weeks (ρ = –0.88, P = 0.002) and is significantly weaker in responders than in non-responders to treatment (P = 0.004) [4]. Furthermore, the operationalization of plasticity significantly differentiated individuals based on their recovery timelines according to the principle that higher plasticity (i.e., weaker connectivity) enables faster transitions to wellbeing and vice versa [5]. Notably, these outcomes hold particular significance for patients with a favorable quality of life, as high plasticity facilitates mental state changes in response to contextual factors [2]. This framework offers a novel mathematical tool to measure plasticity, likely generalizable across levels of analysis and research fields.