Biology subjects

Raveh, B.

Publications and source records attributed to Raveh, B..

10 recordsLinked to original sources

bioRxiv2026

Learning interpretable kinetic models for biomolecular interaction networks

Many cellular machines operate through weak, transient, and multivalent interactions whose functional states are governed by recurring interaction patterns rather than persistent molecular geometries. Here, we introduce interaction-based Markov state models (iMSMs), which construct interpretable kinetic models using unsupervised clustering of time-averaged, identity-resolved interaction distributions around a focal entity. Applied to nucleocytoplasmic transport at two molecular resolutions, iMSMs resolve graded interaction states spanning strong, partial, and weak engagement. During pore transport, partially engaged states provide faster routes to disengagement than strongly bound states, while the networks reveal transport pathways, interaction hubs, bottlenecks, and kinetic commitment. At a finer resolution, FG motifs exchange contacts while remaining associated, recovering established slide-and-exchange dynamics. iMSMs reproduce free energies, permeabilities, and multi-step kinetics, with nearly fourfold faster permeability convergence from truncated trajectories than direct transport-event counting. Thus, iMSMs connect rapidly exchanging contacts to graded interaction states and their kinetics across molecular resolutions.

bioinformatics↗

bioRxiv2026

CliPepPI: Scalable prediction of domain-peptide specificityusing contrastive learning

Domain-peptide interactions mediate a significant fraction of cellular protein networks, yet accurately predicting their specificity remains challenging. Peptide motifs typically have short, fuzzy sequence profiles, and their interactions are often weak and transient, limiting the size, coverage, and quality of experimentally validated domain-peptide datasets. Since true non-binders are rarely known, constructing negative examples often introduces bias. While structure-based prediction methods can achieve high accuracy, they are computationally demanding and difficult to scale to the proteome level. We introduce CLIPepPI, a dual-encoder model that leverages contrastive learning to embed domains and peptides into a shared space directly from sequence. Both encoders are initialized from a protein language model (ESM-C) and fine-tuned using lightweight LoRA adapters, enabling parameter-efficient training on positive pairs alone. To overcome data scarcity, we augment [~]3K protein-peptide complexes from PPI3D with [~]150K domain-peptide pairs derived from protein-protein interfaces. CLIPepPI further injects structural information by marking interface residues in the domain sequence, thus guiding the encoders toward binding regions and linking sequence-level learning with structural context. Competitive performance is achieved across three independent benchmarks: domain-peptide complexes from PPI3D, large-scale phage-library data from ProP-PD, and a curated dataset of nuclear export signal (NES) sequences. We demonstrate scalability and generalization through two applications: (i) proteome-wide NES scanning, and (ii) variant-effect prediction, where score changes in domain-peptide interactions between wild-type and mutant sequences discriminate pathogenic from benign variants. Together, CLIPepPI offers a scalable, structure-informed model for predicting domain-peptide specificity and generating meaningful embeddings suited for large-scale proteomic analyses. CLIPepPI is available at: https://bio3d.cs.huji.ac.il/webserver/clipeppi/.

bioinformatics↗

bioRxiv2025

DeepDynamics resolves cell-subtype and clinicopathological dynamics from bulk RNA-seq to identify mediators of Alzheimer's disease risk

Processes such as Alzheimers disease and aging are shaped by dynamic cascades of cellular and molecular changes. Mapping how genetic and environmental risk factors alter these cascades is critical for pinpointing disease susceptibility and guiding therapeutic intervention, yet this task requires dense sampling of cell-subtype and -subpopulation changes throughout the full course of disease. Bulk measurements provide scale but obscure cell-type and subpopulation resolution, whereas single-cell assays offer resolution but lack sufficient sample size and temporal coverage. We present DeepDynamics, a deep-learning framework that combines a smaller reference sc/snRNA-seq atlas with a larger bulk RNA-seq cohorts to infer cell-subpopulation and clinicopathological dynamics along annotated trajectories. Applied to Alzheimers disease (AD), DeepDynamics mapped 1,092 cortical bulk profiles onto aging and disease trajectories, revealing faster accumulation of AD clinicopathologies, disease-associated glia, and specific neuronal subpopulations in APOE4 carriers. These changes, stronger in females, were accompanied by upregulation of cell-type-specific molecular pathways and early downregulation of heat-shock proteins across cell types. By unifying genetic risk, pathology, and cell-subpopulation dynamics, DeepDynamics provides a scalable approach to prioritize putative mediators of risk and resilience amid the myriad changes along the cascade, thus informing targeted therapies.

bioinformatics↗

bioRxiv2024

Bayesian Occam's Razor to Optimize Models for Complex Systems

Modeling complex biological systems necessitates the integration of vast and multifaceted information spanning various aspects of these systems, and is expected to yield more insights into the system than any of the inputs. Metamodeling, a specialized form of integrative modeling, addresses this by integrating existing models. Developing and optimizing metamodels pose challenges due to the complexities introduced by diverse input models and their inherent uncertainties. In this study, we employ Bayesian formalism to rigorously analyze the propagation of probability throughout the metamodeling process and propose quantitative assessments for it. Building on this, we introduce a method for optimizing metamodeling that adheres to the Bayesian Occams razor rationale, by (i) minimizing model uncertainty; (ii) maximizing model consistency; and (iii) reducing model complexity. To illustrate the benefits of this method, we apply it to the dynamic system of glucose-stimulated insulin secretion in pancreatic {beta}-cells. The optimized metamodel delivers more accurate estimates of impaired {beta}-cell dynamics and function in T2D subjects compared to the non-optimized one, underscoring the critical role of optimization in enhancing both model reliability and applicability. This method is implemented through the Integrative Modeling Platform (IMP), facilitating the development of accurate, precise, and sufficiently simple models for a variety of complex systems.

bioinformatics↗

bioRxiv2024

Interpreting single-cell and spatial omics data using deep networks training dynamics

Single-cell and spatial genomics datasets can be organized and interpreted by annotating single cells to distinct types, states, locations, or phenotypes. However, cell annotations are inherently ambiguous, as discrete labels with subjective interpretations are assigned to heterogeneous cell populations based on noisy, sparse, and high-dimensional data. Here, we show that incongruencies between cells and their input annotations can be identified by analyzing a rich but overlooked source of information: the difficulty of training a deep neural network to assign each cell to its input annotation, or annotation trainability. Furthermore, we demonstrate that annotation trainability encodes meaningful biological signals. Based on this observation, we introduce the concept of signal-aware graph embedding, which facilitates downstream analysis of diverse biological signals in single-cell and spatial omics data, such as the identification of cellular communities corresponding to a target signal. We developed Annotatability, a publicly-available implementation of annotation-trainability analysis. We address key challenges in the interpretation of genomic data, demonstrated over seven single-cell RNA-sequencing and spatial omics datasets, including auditing and rectifying erroneous cell annotations, identifying intermediate cell states, delineating complex temporal trajectories along development, characterizing cell diversity in diseased tissue, identifying disease-related genes, assessing treatment effectiveness, and identifying rare healthy-like cell populations. These results underscore the broad applicability of annotation-trainability analysis via Annotatability for unraveling cellular diversity and interpreting collective cell behaviors in health and disease.

bioinformatics↗

bioRxiv2024

Integrative spatiotemporal map of nucleocytoplasmic transport

Nuclear Pore Complexes (NPCs) enable rapid, selective, and robust nucleocytoplasmic transport. To explain how transport emerges from the system components and their interactions, we used experimental data and theoretical information to construct an integrative Brownian dynamics model of transport through an NPC, coupled to a kinetic model of transport in the cell. The model recapitulates key aspects of transport for a wide range of molecular cargos, including pre-ribosomes and viral capsids. It quantifies how flexible phenylalanine-glycine (FG) repeat proteins raise an entropy barrier to passive diffusion and how this barrier is selectively lowered in facilitated diffusion by the many transient interactions of nuclear transport receptors with the FG repeats. Selective transport is enhanced by "fuzzy" multivalent interactions, redundant FG repeats, coupling to the energy-dependent RanGTP concentration gradient, and exponential dependence of transport kinetics on the transport barrier. Our model will facilitate rational modulation of the NPC and its artificial mimics.

cell biology↗

bioRxiv2023

Dynamic molecular mechanism of the nuclear pore complex permeability barrier

Nuclear pore complexes (NPCs) mediate nucleocytoplasmic transport of specific macromolecules while impeding the exchange of unsolicited material. However, key aspects of this gating mechanism remain controversial. To address this issue, we determined the nanoscopic behavior of the permeability barrier directly within yeast S. cerevisiae NPCs at transport-relevant timescales. We show that the large intrinsically disordered domains of phenylalanine-glycine repeat nucleoporins (FG Nups) exhibit highly dynamic fluctuations to create transient voids in the permeability barrier that continuously shape-shift and reseal, resembling a radial polymer brush. Together with cargo-carrying transport factors the FG domains form a feature called the central plug, which is also highly dynamic. Remarkably, NPC mutants with longer FG domains show interweaving meshwork-like behavior that attenuates nucleocytoplasmic transport in vivo. Importantly, the bona fide nanoscale NPC behaviors and morphologies are not recapitulated by in vitro FG domain hydrogels. NPCs also exclude self-assembling FG domain condensates in vivo, thereby indicating that the permeability barrier is not generated by a self-assembling phase condensate, but rather is largely a polymer brush, organized by the NPC scaffold, whose dynamic gating selectivity is strongly enhanced by the presence of transport factors.

biophysics↗

bioRxiv2021

Bayesian Metamodeling of pancreatic islet architecture and functional dynamics

The pancreatic islet (islet of Langerhans) is a mini-organ comprising several thousand endocrine cells, functioning jointly to maintain normoglycemia. Cellular networks within an islet were shown to influence its function in health and disease, but there are major gaps in our quantitative understanding of such architecture-function relations. Comprehensive modeling of an islet architecture and function requires the integration of vast amounts of information obtained through different experimental and theoretical approaches. To address this challenge, our lab has recently developed Bayesian metamodeling, a general approach for modeling complex systems by integrating heterogeneous input models. Here, we further developed metamodeling and applied it to construct a metamodel of a pancreatic islet. The metamodel relates islet architecture and function by combining a Monte-Carlo model of architecture trained on islet imaging data; and an ordinary differential equations (ODEs) mathematical model of function trained on calcium imaging, hormone imaging, and electrophysiological data. These input models are converted to a standardized statistical representation relying on Probabilistic Graphical Models; coupled by modeling their mutual relations with the physical world; and finally, harmonized through backpropagation. We validate the metamodel using existing data and use it to derive a testable hypothesis regarding the functional effect of varying intercellular connections. Since metamodeling currently requires substantial expert intervention, we also develop an automation tool for converting models to PGMs (step I) using feedforward neural networks. This automation is a first step towards automating the entire metamodeling process, working towards collaborative science through sharing of expertise, resources, data, and models.

systems biology↗

bioRxiv2021

Mechanosensitivity of nucleocytoplasmic transport

Mechanical force controls fundamental cellular processes in health and disease, and increasing evidence shows that the nucleus both experiences and senses applied forces. Here we show that nuclear forces differentially control passive and facilitated nucleocytoplasmic transport, setting the rules for the mechanosensitivity of shuttling proteins. We demonstrate that nuclear force increases permeability across nuclear pore complexes, with a dependence on molecular weight that is stronger for passive than facilitated diffusion. Due to this differential effect, force leads to the translocation into or out of the nucleus of cargoes within a given range of molecular weight and affinity for nuclear transport receptors. Further, we show that the mechanosensitivity of several transcriptional regulators can be both explained by this mechanism, and engineered exogenously by introducing appropriate nuclear localization signals. Our work sets a novel framework to understand mechanically induced signalling, with potential general applicability across signalling pathways and pathophysiological scenarios. One sentence summaryForce application to the nucleus leads to nuclear accumulation of proteins by differentially affecting passive versus facilitated nucleocytoplasmic transport.

biophysics↗

bioRxiv2021

Bayesian metamodeling of complex biological systems across varying representations

Comprehensive modeling of a whole cell requires an integration of vast amounts of information on various aspects of the cell and its parts. To divide-and-conquer this task, we introduce Bayesian metamodeling, a general approach to modeling complex systems by integrating a collection of heterogeneous input models. Each input model can in principle be based on any type of data and can describe a different aspect of the modeled system using any mathematical representation, scale, and level of granularity. These input models are (i) converted to a standardized statistical representation relying on Probabilistic Graphical Models, (ii) coupled by modeling their mutual relations with the physical world, and (iii) finally harmonized with respect to each other. To illustrate Bayesian metamodeling, we provide a proof-of-principle metamodel of glucose-stimulated insulin secretion by human pancreatic {beta}-cells. The input models include a coarse-grained spatiotemporal simulation of insulin vesicle trafficking, docking, and exocytosis; a molecular network model of glucose-stimulated insulin secretion signaling; a network model of insulin metabolism; a structural model of glucagon-like peptide-1 receptor activation; a linear model of a pancreatic cell population; and ordinary differential equations for systemic postprandial insulin response. Metamodeling benefits from decentralized computing, while often producing a more accurate, precise, and complete model that contextualizes input models as well as resolves conflicting information. We anticipate Bayesian metamodeling will facilitate collaborative science by providing a framework for sharing expertise, resources, data, and models, as exemplified by the Pancreatic {beta}-Cell Consortium. Significance StatementCells are the basic units of life, yet their architecture and function remain to be fully characterized. This work describes Bayesian metamodeling, a modeling approach that divides-and-conquers a large problem of modeling numerous aspects of the cell into computing a number of smaller models of different types, followed by assembling these models into a complete map of the cell. Metamodeling enables a facile collaboration of multiple research groups and communities, thus maximizing the sharing of expertise, resources, data, and models. A proof-of-principle is provided by a model of glucose-stimulated insulin secretion produced by the Pancreatic {beta}-Cell Consortium.

cell biology↗