The fundamental understanding of how cells physically interact with each other and their environment is key to understanding their organisation in living tissues. Over the past decades several computational methods have been developed to decipher emergent multi-cellular behaviors. In particular agent-based (or cell-based) models that consider the individual cell as basic modeling unit tracked in space and time enjoy increasing interest across scientific communities. In this article we explore a particular class of cell-based models, so-called Deformable Cell Models (DCMs), that allow to simulate the biophysics of the cell with high realism. After situating this model among other model types, We give an overview of past and recent DCM developments and discuss new simulation results of several applications covering in-vitro and in-vivo systems. Our goal is to demonstrate how such models can generate quantitative added value in biological and biotechnological problems.

Understanding the population dynamics of ecological systems is crucial for predicting shifts in biodiversity and ensuring the protection of these systems. Established models often focus on pairwise species interactions, yet recent studies have highlighted the importance of higher-order interactions (HOIs) in shaping community structure and function. In this study, we investigate the effects of HOIs in an agent-based model with three species engaged in intransitive competition. We introduce an HOI where one species modifies the competition between the other two. We explore the impact of the strength, radius of influence, and speed of this interaction modification on species abundances and oscillations thereof. Our results show that these abundances are not only greatly impacted by the strength, but also by the radius and speed of the interaction modification. A deeper investigation demonstrates that the changes in the oscillations are caused by the interaction modification itself, and not the change in pairwise interaction strength caused by the HOI. These results emphasize the importance of considering the spatio-temporal scales of higher-order interactions when assessing ecosystem stability, highlighting that such interactions can introduce complex dynamical behaviors that go beyond the predictions of traditional pairwise or simpler higher-order models

Human cognition exhibits strong circadian modulation, yet its influence on high-dimensional semantic behavior remains poorly understood. Using large-scale Reddit data, we quantify time-of-day variation in language use by embedding text into a pretrained transformer model and measuring semantic entropy as an index of linguistic exploration-exploitation, for which we show a robust circadian rhythmicity that could be entrained by seasonal light cues. Distinguishing between local and global semantic entropy reveals a systematic temporal dissociation: local semantic exploration peaks in the morning, reflecting broader exploration of semantic space, whereas global semantic diversity peaks later in the day as submissions accumulate around already established topics, consistent with "rich-get-richer" dynamics. These patterns are not explained by sentiment or affective valence, indicating that semantic exploration captures a cognitive dimension distinct from mood. The observed temporal structure aligns with known diurnal patterns in neuromodulatory systems, suggesting that biological circadian rhythms extend to the semantic domain.

Cortical gamma ($γ$)-band activity reflects local excitation-inhibition (E/I) balance. In schizophrenia (SCZ), reduced task-evoked gamma suggests altered E/I dynamics, but it is unclear whether differences stem from input properties or systematic shifts in E/I operating point and gain. We coupled a cochlear-inspired speech front end to a Wilson-Cowan E/I model to simulate gamma responses across three conditions: Healthy, SCZ-speech, and SCZ-semantics. Metrics included event-related spectral perturbation (ERSP$_γ$) and threshold-time fraction ($γ%$). A stable hierarchy emerged: Healthy(speech/semantics) $>$ SCZ(speech) $>$ SCZ(semantics), robust under equal-energy control and gain perturbations. Network dynamics coincided with single-node solutions, supporting interpretability. Pharmacological analogs showed bidirectional effects: reduced inhibition lowered $γ$, while reduced excitation increased $γ$, with no self-sustained oscillations. Findings indicate SCZ gamma deficits align more with shifts in E/I operating point and gain than input differences. This pipeline provides a testable, reusable mechanistic framework for speech-evoked gamma and a baseline for cross-population studies.

This paper investigates the classification capability of small-scale spiking neural networks based on the Leaky Integrate-and-Fire (LIF) neuron model. We analyze the relationship between classification accuracy and three factors: the number of neurons, the number of stimulus nodes, and the number of classification categories. Notably, we employ a large language model (LLM) to assist in discovering the underlying functional relationships among these variables, and compare its performance against traditional methods such as linear and polynomial fitting. Experimental results show that classification accuracy follows a power-law scaling primarily with the number of categories, while the effects of neuron count and stimulus nodes are relatively minor. A key advantage of the LLM-based approach is its ability to propose plausible functional forms beyond pre-defined equation templates, often leading to more concise or accurate mathematical descriptions of the observed scaling laws. This finding has important implications for understanding efficient computation in biological neural systems and for pioneering new paradigms in AI-aided scientific discovery.

In the wake of the SARS-CoV-2 pandemic, there has been heightened interest from applied mathematicians in infectious disease modelling. Modelling efforts often focus on predicting whether diseases are likely to be eliminated or, instead, (re-)emerge, especially as a result of control measures.This tipping point between elimination and infection waves has been successfully anticipated in the literature through the use of early warning signals and such signals often rely on the theory of critical slowing down. Recent developments have shown that these signals (increases in fluctuation variance and return time) can emerge from the system geometry in the case of non-normal dynamics rather than a change in asymptotic stability. We show how such dynamical behaviour occurs in the fluctuations from the mean-field in general stochastic systems. Using the susceptible-infectious-recovered model as an example application, we analyse how critical-like behaviour can be exploited to anticipate infection waves in the absence of an equilibrium bifurcation.

Supervised deep learning models often achieve excellent performance within their training distribution but struggle to generalize beyond it. In cancer histopathology, for example, a convolutional neural network (CNN) may classify cancer severity accurately for cancer types represented in its training data, yet fail on related but unseen types. Although adenocarcinomas from different organs share morphological features that might support limited cross-domain generalization, addressing domain shift directly is necessary for robust performance. Domain adaptation offers a way to transfer knowledge from labeled data in one cancer type to unlabeled data in another, helping mitigate the scarcity of annotated medical images. This work evaluates cross-domain classification performance among lung, colon, breast, and kidney adenocarcinomas. A ResNet50 trained on any single adenocarcinoma achieves over 98% accuracy on its own domain but shows minimal generalization to others. Ensembling multiple supervised models does not resolve this limitation. In contrast, converting the ResNet50 into a domain adversarial neural network (DANN) substantially improves performance on unlabeled target domains. A DANN trained on labeled breast and colon data and adapted to unlabeled lung data reaches 95.56% accuracy. We also examine the impact of stain normalization on domain adaptation. Its effects vary by target domain: for lung, accuracy drops from 95.56% to 66.60%, while for breast and colon targets, stain normalization boosts accuracy from 49.22% to 81.29% and from 78.48% to 83.36%, respectively. Finally, using Integrated Gradients reveals that DANNs consistently attribute importance to biologically meaningful regions such as densely packed nuclei, indicating that the model learns clinically relevant features and can apply them to unlabeled cancer types.

Contact tracing (CT) plays a pivotal role in controlling early epidemic spread, particularly when a novel infectious disease emerges. However, the quantitative impact of missing information -- such as untraced cases or unnotified contacts -- on the effectiveness of CT remains insufficiently understood. Using a stochastic agent-based model with sociodemographics from metropolitan areas in South Korea, we simulate how different forms of information loss affect epidemic spreading dynamics. We construct information-loss scenarios based on two types: infector-omission (IO) and contact-omission (CO), including selective (SCO) and uniform (UCO) scenarios; IO corresponds to the omission of infected individuals (nodes) from the tracing process, leading to the loss of all movement trajectories and downstream transmission links originating from them, whereas CO corresponds to the omission of specific contact events (edges), in which infected individuals are identified but some of their transmission links fail to be detected or notified. The sensitivity of epidemic dynamics to increasing omission rates differs markedly between the two types: IO scenarios exhibit substantially stronger and more abrupt changes in transmission structure and epidemic outcomes, whereas CO scenarios produce more gradual effects. In both scenarios, the magnitude of these effects varies across cities, with a lower-population city (Busan) showing greater tolerance to information loss than the largest city (Seoul), underscoring the importance of regional tailoring in CT strategies. Both IO and CO scenarios also lead to an increase in the transmission network diameter as information loss grows, indicating that a small network diameter reflects effective contact tracing that limits the depth of transmission chains.

The rapid development of high-throughput sequencing technologies has led to an explosive increase in biological sequence data, making sequence clustering a fundamental task in large-scale bioinformatics analyses. Unlike traditional clustering problems, biological sequence clustering faces unique challenges due to the lack of direct similarity measures, strict biological constraints, and demanding requirements for both scalability and accuracy. Over the past decades, a wide variety of methods have been developed, differing in how they model sequence similarity, construct clusters, and prioritize optimization objectives. In this review, we provide a comprehensive methodological overview of biological sequence clustering algorithms. We begin by summarizing the main strategies for modeling sequence similarity, which can be divided into three stages: sequence encoding, feature generation, and similarity measurement. Next, we discuss the major clustering paradigms, including greedy incremental, hierarchical, graph-based, model-based, partitional, and deep learning approaches, highlighting their methodological characteristics and practical trade-offs. We then discuss clustering objectives from three key perspectives: scalability and resource efficiency, biological interpretability, and robustness and clustering quality. Organizing existing methods along these dimensions allows us to explore the trade-offs in biological sequence clustering and clarify the contexts in which different approaches are most appropriate. Finally, we identify current limitations and challenges, providing guidance for researchers and directions for future method development.

Excessive consumption of dietary sugars is a major contributor to metabolic disorders, driving global interest in finding alternative sweeteners with reduced caloric impact. Natural sweet proteins, such as brazzein, offer exceptional sweetness intensity with little caloric contribution. However, their widespread use is limited by restricted natural diversity, low stability, and high production costs. Recent advances in structural biology and de novo protein design provide new opportunities to overcome these limitations through rational engineering. In this study, we report an integrated computational pipeline for the de novo design of protein binders targeting the human sweet taste receptor subunit TAS1R2, a key component of the heterodimeric class C G protein-coupled receptor mediating sweetness perception. The workflow combines diffusion-based backbone generation (RFdiffusion), neural network-guided sequence design (ProteinMPNN), structure-based filtering using Boltz-1, and binding energy evaluation via MM/GBSA calculations. Using the recently resolved cryo-EM structure of the TAS1R2 receptor, protein binders were designed to target both the Venus Flytrap Domain and the cysteine-rich domain of TAS1R2. A few designed binders exhibited favorable structural confidence and predicted binding energetics. In particular, Binder2 exhibited brazzein-like structural plausibility through specific short-range CRD contacts, while Binder1 displayed the strongest predicted binding affinity. Structural analyses of the binder-receptor complex revealed distinct binding modes and secondary structure profiles among the designs. This study demonstrates the feasibility of de novo designing protein binders that emulate key functional properties of natural sweet proteins, establishing a computational framework for the rational development of next-generation protein-based sweeteners.

Omics data, such as transcriptomics, proteomics, and metabolomics, provide critical insights into disease mechanisms and clinical outcomes. However, their high dimensionality, small sample sizes, and intricate biological networks pose major challenges for reliable prediction and meaningful interpretation. Graph Neural Networks (GNNs) offer a promising way to integrate prior knowledge by encoding feature relationships as graphs. Yet, existing methods typically rely solely on either an externally curated feature graph or a data-driven generated one, which limits their ability to capture complementary information. To address this, we propose the external and generated Graph Neural Network (engGNN), a dual-graph framework that jointly leverages both external known biological networks and data-driven generated graphs. Specifically, engGNN constructs a biologically informed undirected feature graph from established network databases and complements it with a directed feature graph derived from tree-ensemble models. This dual-graph design produces more comprehensive embeddings, thereby improving predictive performance and interpretability. Through extensive simulations and real-world applications to gene expression data, engGNN consistently outperforms state-of-the-art baselines. Beyond classification, engGNN provides interpretable feature importance scores that facilitate biologically meaningful discoveries, such as pathway enrichment analysis. Taken together, these results highlight engGNN as a robust, flexible, and interpretable framework for disease classification and biomarker discovery in high-dimensional omics contexts.

Probabilistic mental simulation is thought to play a key role in human reasoning, planning, and prediction, yet the demands of simulation in complex environments exceed realistic human capacity limits. A theory with growing evidence is that people simulate using simplified representations of the environment that abstract away from irrelevant details, but it is unclear how people determine these simplifications efficiently. Here, we present a "Just-in-Time" framework for simulation-based reasoning that demonstrates how such representations can be constructed online with minimal added computation. The model uses a tight interleaving of simulation, visual search, and representation modification, with the current simulation guiding where to look and visual search flagging objects that should be encoded for subsequent simulation. Despite only ever encoding a small subset of objects, the model makes high-utility predictions. We find strong empirical support for this account over alternative models in a grid-world planning task and a physical reasoning task across a range of behavioral measures. Together, these results offer a concrete algorithmic account of how people construct reduced representations to support efficient mental simulation.

Developing epithelial tissues coordinate cell proliferation and mechanical forces to achieve proper size and shape. As epithelial cells tightly adhere together to form the confluent tissue, the distribution of cell areas significantly influences possible patterns of cellular packing and thereby also the mechanics of the epithelium. Therefore, it is important to understand the origin of cell area heterogeneity in developing tissues and, if possible, how to control it. Previous models of cell growth and division have been successful in accounting for experimentally observed area distributions in cultured cells and bacterial colonies, but developing tissues present additional complexity due to self-organized patterns of mechanical stresses that guide morphogenesis. Here, we address this challenge focusing on the D. melanogaster wing disc epithelium. We consider a simple model that couples cell cycle dynamics to tissue mechanics. From time-lapse imaging of the cellular network, we extract all model parameters - cell growth rates, division rates, and mechanical fluctuations - revealing that they all depend on cell size. With these independently measured parameters, our model quantitatively reproduces the observed cell area distribution without any fitting parameters and further predicts tissue pressure gradients, in quantitative agreement with previously published data. Importantly, we find that cell proliferation accounts for 85% of cell area variance, establishing it as the dominant source of packing disorder that influences tissue mechanics and organization.

The use of multi-centric analyses is crucial for obtaining sufficient sample sizes and representative clinical populations in experimental studies. In this setting, data harmonization techniques are typically employed to address systematic biases and ensure the interoperability of the data. State-of-the-art harmonisation approaches are based on the statistical theory of random effect modeling, allowing to account for either linear of non-linear biases and batch effects. However, optimizing these statistical methods generally requires data centralization at some point during the analysis pipeline, therefore introducing the risk of exposing individual patient information while posing significant data governance issues. To overcome this challenge, in this paper we present Fed-ComBat, a federated framework for batch effect harmonization on decentralized data. Fed-ComBat enables the preservation of nonlinear covariate effects without requiring centralization of data and without prior parametric hypothesis on the variables to account for. We demonstrate the effectiveness of Fed-ComBat against a comprehensive panel of existing approaches based on the state-of-the-art ComBat, along with distributed and nonlinear variants. Our experiments are based on extensive simulated data, and on the analysis of multiple cohorts based on 7 neuroimaging studies comprising healthy controls (CI) and subjects with various disorders such as Parkinson's disease (PD), Alzheimer's disease (AD), and autism spectrum disorder (ASD). Our results show that in a federated settings, Fed-ComBat harmonization exhibits comparable results to centralized methods for both linear and nonlinear cases. On real data, harmonized trajectories of the thickness ofthe right hippocampus across lifespan measured on a set of 7 public studies show comparable results between centralized and federated models and are consistent with the literature when using a nonlinear model. The code is publicly available at: https://gitlab.inria.fr/greguig/fedcombat

Background: Being born very or extreme preterm is a major source of cerebral anomalies and neurodevelopmental disorders, whose occurrence depends on many perinatal factors. A better understanding of these factors could be provided by cerebral Magnetic Resonance Imaging (MRI) at term-equivalent age (TEA). Objective: To investigate, through cerebral TEA-MRIs, the relationship between the main perinatal factors, the occurrence of cerebral anomalies, and cerebral volumetry. Methods: We assembled a cohort of preterm babies born before 32 weeks of gestation who underwent a cerebral TEA-MRI. We assessed cerebral anomalies using a radiological scoring system -- the Kidokoro scoring -- and performed cerebral volumetry. We investigated the relationships between 9 clinical factors (birth characteristics, resuscitation treatments{\ldots}), Kidokoro scores, and brain volumes. Results: Among 110 preterms who underwent a cerebral MRI at TEA, only 6% suffered moderate-to-severe brain anomalies. We identified mechanical ventilation as a risk factor for cerebral anomalies (adjusted Odds-Ratio aOR = 4.6, 95% Confidence Interval CI [1.7-12.8]) and prolonged parenteral nutrition as a protective factor for white matter anomalies (aOR = 0.2, 95%CI [0.1-0.8]). Mechanical ventilation (p = 0.01) and being born small for gestational age (p < 0.001) were risk factors for the reduction of cerebral volumes. An increase in brain lesion severity was associated with decreased cerebral volumes (p = 0.016). Conclusion: Our study highlights the importance of treatment-related perinatal factors on the occurrence of cerebral anomalies in very and extreme preterms, and the interest in using both qualitative (Kidokoro scoring) and quantitative (volumetry) MRI-tools.

Light driven oxygen formation in Photosystem II protein is a fundamental process that sustains our biosphere and serves as a blue print to future clean energy solutions due to its high energy conversion efficiency. Last decade of intense research by advanced physical techniques delivered new insights on the structure and function of the Mn4CaO5 cluster a center of the oxygen evolving complex (OEC). However, discrepancies in experimental observations and computational models persist impeding the understanding of the O-O bond formation and the role of the protein environment in the process. Here we show that i) assignment of the OEC unique oxygen O3 ligated by histidine (His337) via dynamic H-bond as a slow exchanging substrate and ii) its coupling with O6 oxygen generated at Mn1 in the S2 to S3 transition give the O-O bond formation mechanism most consistent with all currently available experimental data. Proposal shows how protein environment can steer the O-O bond formation by charge control via H-bond and open coordination of Mn1. Obtained O3-O6 peroxide is at lower energy than peroxides in the most studied O5-O6 bond formation pathway. His337 appears to be similar to distal His in globins used for management of the O2 and H2O2 intermediates. The new mechanism breaks the prior impasse and will undoubtedly invigorate future detailed studies uncovering its further details.

Accurate segmentation of brain tumors is essential for clinical diagnosis and treatment planning. Deep learning is currently the state-of-the-art for brain tumor segmentation, yet it requires either large datasets or extensive computational resources that are inaccessible in most areas. This makes the problem increasingly difficult: state-of-the-art models use thousands of training cases and vast computational power, where performance drops sharply when either is limited. The top performer in the Brats GLI 2023 competition relied on supercomputers trained on over 92,000 augmented MRI scans using an AMD EPYC 7402 CPU, six NVIDIA RTX 6000 GPUs (48GB VRAM each), and 1024GB of RAM over multiple weeks. To address this, the Karhunen--Loève Expansion (KLE) was implemented as a feature extraction step on downsampled, z-score normalized MRI volumes. Each 240$\times$240$\times$155 multi-modal scan is reduced to four $48^3$ channels and compressed into 32 KL coefficients. The resulting approximate reconstruction enables a residual-based anomaly map, which is upsampled and added as a fifth channel to a compact 3D U-Net. All experiments were run on a consumer workstation (AMD Ryzen 5 7600X CPU, RTX 4060Ti (8GB VRAM), and 64GB RAM while using far fewer training cases. This model achieves post-processed Dice scores of 0.929 (WT), 0.856 (TC), and 0.821 (ET), with HD95 distances of 2.93, 6.78, and 10.35 voxels. These results are significantly better than the winning BraTS 2023 methodology for HD95 distances and WT dice scores. This demonstrates that a KLE-based residual anomaly map can dramatically reduce computational cost and data requirements while retaining state-of-the-art performance.

The digitization of biological specimens has revolutionized the field of morphology, creating large collections of 3D data, and microCT in particular. This revolution was initially supported by the development of open-source software tools, specifically the development of SlicerMorph extension to the open-source image analytics platform 3D Slicer. Through SlicerMorph and 3D Slicer, biologists, morphologists and scientists in related fields have all the necessary tools to import, visualize and analyze these complex and large datasets in a single platform that is flexible and expandible, without the need of proprietary software that hinders scientific collaboration and sharing. Yet, a significant "compute gap" remains: While data and software are now open and accessible, the necessary high-end computing resources to run them are often not equally accessible in all institutions, and particularly lacking at Primarily Undergraduate Institutions (PUIs) and other educational settings. Here, we present MorphoCloud, an "IssuesOps"-based platform that leverages Github Actions and the JetStream2 cloud farm to provide on-demand, research-grade computing environments to researchers working with 3D morphological datasets. By delivering a GPU-accelerated full desktop experience via a web browser, MorphoCloud eliminates hardware barriers, enabling complex 3D analysis and AI-assisted segmentation. This paper explains the platform and its architecture, as well as use cases it is designed to support.

Object binding, the brain's ability to bind the many features that collectively represent an object into a coherent whole, is central to human cognition. It groups low-level perceptual features into high-level object representations, stores those objects efficiently and compositionally in memory, and supports human reasoning about individual object instances. While prior work often imposes object-centric attention (e.g., Slot Attention) explicitly to probe these benefits, it remains unclear whether this ability naturally emerges in pre-trained Vision Transformers (ViTs). Intuitively, they could: recognizing which patches belong to the same object should be useful for downstream prediction and thus guide attention. Motivated by the quadratic nature of self-attention, we hypothesize that ViTs represent whether two patches belong to the same object, a property we term IsSameObject. We decode IsSameObject from patch embeddings across ViT layers using a quadratic similarity probe, which reaches over 90% accuracy. Crucially, this object-binding capability emerges reliably in DINO, CLIP, and ImageNet-supervised ViTs, but is markedly weaker in MAE, suggesting that binding is not a trivial architectural artifact, but an ability acquired through specific pretraining objectives. We further discover that IsSameObject is encoded in a low-dimensional subspace on top of object features, and that this signal actively guides attention. Ablating IsSameObject from model activations degrades downstream performance and works against the learning objective, implying that emergent object binding naturally serves the pretraining objective. Our findings challenge the view that ViTs lack object binding and highlight how symbolic knowledge of "which parts belong together" emerges naturally in a connectionist system.

If a CT scanner introduces a new artifact not present in the training labels, the model may misclassify the images. Although modern CT scanners include design features which mitigate these artifacts, unanticipated or difficult-to-mitigate artifacts can still appear in practice. The direct solution of labeling images from this new distribution can be costly. As a more accessible alternative, this study evaluates domain adaptation as an approach for training models that maintain classification performance despite new artifacts, even without corresponding labels. We simulate ring artifacts from detector gain error in sinogram space and evaluate domain adversarial neural networks (DANN) against baseline and augmentation-based approaches on the OrganAMNIST abdominal CT dataset. We simulate the absence of labels from an unseen distribution via masking in the loss function and selectively detaching unlabeled instances from the computational graph. Our results demonstrate that baseline models trained only on clean images fail to generalize to images with ring artifacts, and traditional augmentation with other distortion types provides no improvement on unseen artifact domains. In contrast, the DANN approach improves classification accuracy on ring artifact images using only unlabeled artifact data during training, demonstrating the viability of domain adaptation for artifact robustness. The domain-adapted model achieved a classification accuracy of 77.4% on ring artifact test data, 38.7% higher than a baseline model only trained on images with no artifact. These findings provide empirical evidence that domain adaptation can effectively address distribution shift in medical imaging without requiring expensive expert labeling of new artifact distributions, suggesting promise for deployment in clinical settings where novel artifacts may emerge.

Population cycles are important components of many natural systems. Most studied in short-lived and small-bodied species, cycles frequently appear to be driven by density-dependent feedbacks. However, compelling evidence of cycles -- often more qualitative than quantitative -- also exists in large mammals. Among ungulates, both density-dependent vital rates and 'cohort effects' (lasting impacts of birth conditions on fecundity and survival) exist, but the implications of such feedbacks for oscillatory population dynamics have not been explored. Here, we present a synthetic model of ungulate population dynamics, parameterized for barren-ground caribou (Rangifer tarandus groenlandicus) and motivated by extensive Indigenous knowledge suggesting decades-long fluctuations in abundance. Caribou herds are theorized to be subject to both cohort effects and density dependence, and we linked these endogenous factors with environmental stochasticity to understand cycling. Using wavelet analysis, we characterized periodic phenomena and performed sensitivity analyses to clarify the drivers and characteristics of population cycles. We found that cohort effects, predominantly those impacting survival, can produce long-period oscillatory behavior across a wide range of environments and demographic structures. Our modeling framework is generalizable to other long-lived, large-bodied species with complex demography, and collectively, these efforts broaden the scope of inquiry into proximal drivers of population cycling.

Neural networks based on soft and biological matter constitute an interesting potential alternative to traditional implementations based on electric circuits. DNA is a particularly promising system in this context due its natural ability to store information. In recent years, researchers have started to construct neural networks that are based on DNA. In this chapter, I provide a very basic introduction to the concept of DNA neural networks, aiming at an audience that is not familiar with biochemistry.

Stochastic processes have found numerous applications in science, as they are broadly used to model a variety of natural phenomena. Due to their intrinsic randomness and uncertainty, they are, however, difficult to characterize. Here, we introduce an unsupervised machine learning approach to determine the minimal set of parameters required to effectively describe the dynamics of a stochastic process. Our method builds upon an extended $β$-variational autoencoder architecture. By means of simulated datasets corresponding to paradigmatic diffusion models, we showcase its effectiveness in extracting the minimal relevant parameters that accurately describe these dynamics. Furthermore, the method enables the generation of new trajectories that faithfully replicate the expected stochastic behavior. Overall, our approach enables the autonomous discovery of unknown parameters describing stochastic processes, hence enhancing our comprehension of complex phenomena across various fields.