Biology subjects

Periwal, V.

Publications and source records attributed to Periwal, V..

14 recordsLinked to original sources

bioRxiv2026

Islet size and spatial organization govern endocrine cell composition throughout the human pancreas

Stem cell-derived (SC) islets hold promise as a transformative therapy for type 1 diabetes (T1D). To enhance graft uniformity, SC-islet organoids are often dissociated and reaggregated into standardized sizes. However, how endocrine cell composition varies with islet size in native human islets remains largely unknown. Defining this relationship is essential for establishing physiological benchmarks for islet engineering for transplantation. In this study, we examined the relationships among islet size, endocrine cell composition, and spatial distribution throughout whole pancreata from 10 age-matched, non-diabetic male subjects. Insulin-, glucagon-, and somatostatin-positive endocrine cells were identified as single-, double-, and triple-hormone populations ranging from non-islet single endocrine cells and small clusters (niSSCs) to large islets. Approximately 80% of the total endocrine area was contained in islets positive for all three hormones. Islet size strongly correlated with endocrine composition. Specifically, larger islets had increased alpha-cell fractions and reduced beta-cell fractions, while delta-cell fractions were consistent across sizes. Spatial analysis revealed that niSSCs, mono-, and tri-hormone islets frequently exhibit a clustered distribution. Additionally, tri-hormone islets were positioned near the core of the pancreas, whereas mono-somatostatin islets were often localized in the periphery. These findings were consistent across all three pancreatic regions (head, body and tail).

pathology↗

bioRxiv2026

Mapping Gene Expression to an Interpretable Semantic Space

Cell embeddings organize single-cell expression data, but their dimensions have no biological meaning, so clusters are interpreted afterward. We present MESIC (Mapping Expression to Semantic space with Interpretable Components), which builds the written knowledge about genes held in curated databases into the dimensions themselves. A biomedical language model converts each gene's summary into a semantic embedding. MESIC compresses these embeddings into a small number of components, each concentrated on a small set of genes and explained by their annotations. The components are computed once from the summaries, so any expression dataset can be mapped onto them, and every cluster, outlier, or cell-type assignment is then characterized by named genes. In cardiomyocytes, outliers in the component space were enriched for hypertrophic cardiomyopathy. In a lung atlas, unsupervised clusters in that space matched the broad cell types that experts had annotated. In both, the components that separated the cells matched their known biology. For about half of the cells that the atlas itself had left unannotated, the same space gave a confident cluster assignment, and with it an interpretation through component-associated genes. Gene summaries thus give single-cell analysis a coordinate system in which every result is traced to genes and what is written about them.

bioinformatics↗

bioRxiv2026

Topological Entropy Correlates with the Predictive Power of Multiplexed Ensemble Reservoir Computing

Modeling nonlinear, multiscale, and transiently chaotic biological processes remains a major challenge in computational biology. Traditional deep learning models, while powerful, require large datasets and lack mechanistic interpretability, limiting their effectiveness for time-resolved biological systems. Reservoir computing (RC) offers a promising alternative by leveraging the rich transient dynamics of fixed nonlinear systems, yet standard RC architectures struggle with high-dimensional biological data and complex temporal regimes. Here, we introduce Dynamical System Machine Learning (DynML), a multiplexed reservoir framework designed to model gene-expression dynamics in systems such as liver regeneration and Drosophila embryogenesis. DynML encodes biological signals using heterogeneous Lorenz reservoirs and employs a single global readout to capture stage-dependent dynamics with high predictive accuracy. We further show that reservoir topological entropy quantitatively predicts model performance, linking dynamical richness to biological forecasting accuracy. Beyond biological time-series modeling, we demonstrate the generality of DynML on the MNIST handwritten digit classification task using a Rossler-based chaotic reservoir, showing that fixed dynamical cores with linear readouts can also support high-dimensional static classification. Overall, DynML provides a scalable, interpretable, and computationally efficient framework that unifies biological time-series modeling and conventional machine-learning tasks within a single dynamical systems paradigm. Author summaryComplex biological phenomena such as development, regeneration, and disease progression emerge from time-dependent gene-expression programs governed by nonlinear, multiscale dynamics. Capturing these dynamics remains challenging for conventional machine-learning approaches, which typically require large datasets and lack interpretability. In this study, we introduce Dynamical System Machine Learning (DynML), a modeling framework that leverages the transient dynamics of chaotic systems to learn and predict biological time series. DynML transforms gene-expression measurements into high-dimensional dynamical representations using ensembles of nonlinear reservoirs, enabling accurate prediction of future expression states with simple and interpretable linear readouts. We apply DynML to both synthetic dynamical systems and real biological datasets, including Drosophila embryonic development and human liver regeneration, where it achieves high predictive accuracy across multiple temporal transitions. Importantly, we show that the predictive performance of DynML is strongly linked to the topological entropy of the reservoir dynamics, providing a principled and quantitative measure of model expressiveness. Beyond biological time-series prediction, we demonstrate that the same dynamical framework can also classify static data, achieving strong performance on handwritten digit recognition. Together, our results establish DynML as a scalable and interpretable approach for modeling complex biological dynamics, and highlight how concepts from dynamical systems theory can guide the design of effective machine-learning models for biological data.

systems biology↗

bioRxiv2025

Hybrid deep learning-mechanistic modeling of cellular dynamics from a spatiotemporal single-cell atlas

Single-cell measurement technologies provide a powerful framework for studying cellular heterogeneity, transitions, and regulatory networks, yet reconstructing the underlying dynamical processes governing these transitions remains a major challenge due to the high dimensionality of gene expression data. To address this, we develop a variational autoencoder (VAE)-latent neural ordinary differential equation (ODE) approach that learns a low-dimensional latent representation of cellular states and models their temporal evolution. We apply our framework to the single-cell fluorescence imaging spatiotemporal atlas of Drosophila melanogaster blastoderm embryos via spatial registration, which comprises six registered developmental time points prior to gastrulation. In our approach, gene expression profiles are encoded into a latent space where a neural ordinary differential equation (ODE) is trained to capture the continuous dynamics of cellular states over time, and a decoder subsequently maps these evolving latent representations back to the original high-dimensional gene expression space, enabling accurate reconstruction of observed transcriptional patterns. While such black-box deep learning approaches excel at capturing complex dynamical trajectories, they are inherently limited in their ability to predict the effects of combinatorial perturbations. To overcome this limitation, we leverage the inferred latent dynamics as a foundation for fitting a mechanistic Hill-function model of gene regulation, which--guided by the black-box representations--enables interpretable predictions by systematically perturbing couplings between specific gene pairs and thus provides mechanistic insight into developmental regulatory programs. Author summaryWe set out to learn how genes interact over time during early development using data measured from individual cells. Many powerful deep learning models can fit complex patterns, but they are hard to edit or use for "what-if" genetic changes. Classical mechanistic models are easier to interpret but struggle when the data are sparse. We built a hybrid approach that combines both strengths. First, we trained a deep learning model to place each cells gene activity into a smooth, low-dimensional space and to estimate how these states change over time. We then used those estimates to fit a simple, editable model of gene regulation based on standard biochemical response curves. This second model lets us turn specific regulatory connections on or off and predict the effects without retraining the whole system. Using a spatiotemporal atlas of the fruit fly embryo, we show that the hybrid model makes accurate short-term predictions and yields clear, testable hypotheses about which genes control others. Our study demonstrates a practical way to turn flexible deep learning into interpretable, mechanistic insight for dynamic single-cell data.

systems biology↗

bioRxiv2025

Forecasting drug resistant HIV protease evolution

Protease inhibitors (PIs) target the protease (PR) enzyme to suppress viral replication. Their efficacy in human immunodeficiency virus treatment is compromised by the emergence of drug-resistant strains. Therefore, forecasting drug-resistance during viral evolution would help in the design of effective treatment strategies. We develop a probabilistic large-deviation model to infer epistatic interactions in genotypes observed in different treatment regimens and compute transition probabilities of point-mutations conditioned on the genotype and the treatment regimen. We simulate stochastic evolutionary paths weighted by such transition probabilities and show that low probable mutations are required for the viral population to evolve to diverse fit genotypes. We train classification models using a clinical data set of in vitro susceptibility tests to learn to infer the drug resistance of a genotype. We infer drug resistance along simulated evolutionary paths and predict that the combination PI-therapy of Atazanavir (ATV) and Ritonavir (RTV) is the least drug resistant. Without prior knowledge of PI-associated mutations, our model predicts known primary and secondary PI-resistant mutations as critical to drug resistance. This validates that our model learned mechanistic relations in the small data sets, tackling the challenge of sparse sequence data compared to the large combinatorial complexity of protein evolution and changing functionality in dynamic environments.

systems biology↗

bioRxiv2025

Impact of low-calorie sweeteners on gut bacteria is modulated by common xenobiotics

The gut microbiota is implicated in adverse effects associated with low-calorie sweeteners. Yet, the direct impact of sweeteners on gut bacteria remains largely uncharacterized. Here we report interactions between 25 phylogenetically diverse gut bacterial strains and 39 commercially used sweeteners. We tested these sweeteners individually and in combination with four commonly co-consumed compounds, viz., advantame, caffeine, vanillin, and duloxetine. Three quarters of the tested sweeteners individually impacted growth of at least one bacterial strain. Further, over 100 interactions were found between sweeteners and the four co-consumed compounds. Isosteviol, a commonly used sweetener, and duloxetine, an antidepressant, synergistically inhibited Roseburia intestinalis, a bacterium previously linked to glucose homeostasis, and Parabacteroides merdae, a prevalent commensal linked to healthy microbiota. Proteomic, metabolomic, and genetic analyses indicate altered small molecule transport underpinning this sweetener-drug synergy. The isosteviol-duloxetine combination also modulated metabolism of a synthetic gut bacterial community leading to increased toxicity to HeLa cells and altered secretion of inflammation modulatory cytokines IL-6 and IL-8 by Caco-2 cells. Together, our data bring forward the prevalence of interactions between low-calorie sweeteners and common xenobiotics.

microbiology↗

bioRxiv2025

Deep learning approach to parameter optimization for physiological models

The inference of nonlinear dynamics and parameters in biological data modeling is challenging. Conventional methodologies, based on hypothetical underlying mechanisms, complicate inference because standard parameter optimization methods are difficult to constrain to biological ranges. Here, we propose a novel method to evaluate and improve putative models using neural networks to simultaneously address biological modeling, parametrization, and parameter inference. As an example, utilizing data from clinical frequently sampled intravenous glucose tolerance testing, we introduce two physiological lipolysis models (with parameters) of the dynamics of glucose, insulin, and free fatty acids (FFA). Parameter values are obtained via optimization from the limited clinical data. We then generate large quantities of simulated data from the model by sampling parameters within physiological ranges. A convolutional neural network is trained to take the simulated data time courses of glucose, insulin, and FFA as input and output the model parameters. The performance of the trained neural network is evaluated for both parameter inference and reconstruction of trajectories over a testing dataset and from optimized model-fitting curves. We show that our methodology enables accurate parameter inference and trajectory reconstruction over the testing dataset and optimized model-fitting curves. The trained neural network produces consistently high R2 values and low p-values across different feature engineering strategies and training dataset sizes. We assess the impact of feature engineering choices and training dataset size on inference performance, demonstrating that appropriately designed feature transformations and certain activation function improve accuracy. Our results establish a deep learning framework for parameter inference in mathematical models, which can be adapted to various physiological systems.

physiology↗

bioRxiv2025

Donor-Specific Digital Twin for Living Donor Liver Transplant Recovery

Liver resection initiates a meticulously coordinated hyperplasia process characterized by regulated cell proliferation that drives liver regeneration. This process concludes with the complete restoration of liver mass, showcasing the precision and robustness of this homeostasis. The remarkable capacity of the liver to regenerate rapidly into a fully functional organ has been crucial to the success of living donor liver transplantation (LDLT). In healthy livers, hepatocytes typically remain in a quiescent state (G0). However, following partial hepatectomy, these cells transition to the G1 phase to re-enter the cell cycle. Surgical resection induces various stresses, including physical injury, altered blood flow, and increased metabolic demands. These all trigger the activation and suppression of numerous genes involved in tissue repair, regeneration, and functional recovery. Both coding and noncoding RNAs detectable in the bloodstream during this process provide valuable insights into the gene responses driving liver recovery. This study integrates clinical gene expression data into a previously developed mathematical model of liver regeneration, which tracks transitions among quiescent, primed, and proliferating hepatocytes to construct virtual, patient-specific liver models. Using whole transcriptome RNA sequencing data from 12 healthy LDLT donors, collected at 14 time points over a year, we identified liver resection-specific gene expression patterns through Weighted Gene Co-expression Network Analysis (WGCNA). These patterns were organized into distinct clusters with unique transcriptional dynamics and mapped to model variables using deep learning techniques. Consequently, we developed a Personalized Progressive Mechanistic Digital Twin (PePMDT) for the livers of LDLT donors. The resulting PePMDT predicts individual patient recovery trajectories by leveraging blood-derived gene expression data to simulate regenerative responses. By transforming gene expression profiles into dynamic model variables, this approach bridges clinical data and mathematical modeling, providing a robust platform for personalized medicine. This study highlights the transformative potential of data-driven frameworks like PePMDT in advancing precision medicine and optimizing recovery outcomes for LDLT donors.

systems biology↗

bioRxiv2025

Sensitivity based model agnostic scalable explanations of deep learning

Deep neural networks (DNNs) are powerful tools for data-driven predictive machine learning, but their complex architecture obscures mechanistic relations that they have learned from data. This information is critical to the scientific method of hypotheses development, experiment design, and model validation, especially when DNNs are used for biological and clinical predictions that affect human health. We design SensX, a model agnostic explainable AI (XAI) framework that outperformed current state-of-the-art XAI in accuracy (up to 52% higher) and computation time (up to 158 times faster), with higher consistency in all cases. It also determines an optimal subset of important input features, reducing dimensionality of further analyses. SensX scaled to explain vision transformer (ViT) models with more than 150, 000 features, which is computationally infeasible for current state-of-the-art XAI. SensX validated that ViT models learned justifiable features as important for different facial attributes of different human faces. SensX revealed biases inherent to the ViT architecture, an observation possible only when importance of each feature is explained. We trained DNNs to annotate biological cell types using single-cell RNA-seq data and SensX determined the sets of genes that the DNNs learned to be important to different cell types.

systems biology↗

bioRxiv2022

Protein Structure Prediction with Expectation Reflection

Sequence covariation in multiple sequence alignments of homologous proteins has been used extensively to obtain insights into protein structure. However, global statistical inference is required in order to ascertain direct relationships between amino acid positions in these sequences that are not simply secondary correlations induced by interactions with a third residue. Methods for statistical inference of such covariation have been developed to exploit the growing availability of sequence data. These hints about the folded protein structure provide critical a priori information for more detailed 3D predictions by neural networks. We present a novel method for protein structure inference using an iterative parameter-free model estimator which uses the formalism of statistical physics. With no tunable learning rate, our method scales to large system sizes while providing improved performance in the regime of small sample sizes. We apply this method to 40974 PDB structures and compare its performance to that of other methods. Our method outperforms existing methods for 76% of analysed proteins.

molecular biology↗

bioRxiv2021

Mechanistic Gene Networks Inferred from Single-Cell Data are Better Predictors than Neural Networks.

AO_SCPLOWBSTRACTC_SCPLOWWith advances in single-cell techniques, measuring gene dynamics at cellular resolution has become practicable. In contrast, the increased complexity of data has made it more challenging computationally to unravel underlying biological mechanisms. Thus, it is critical to develop novel computational methods capable of dealing with such complexity and of providing predictive deductions from such data. Many methods have been developed to address such challenges, each with its own advantages and limitations. We present an iterative regression algorithm for inferring a mechanistic gene network from single-cell data. Using this regression, we infer a developmental model for the gene dynamics in Drosophila melanogaster blastoderm embryo. Our results show that the predictive power of the inferred model is higher than that of other models inferred with least squares and ridge regressions. Furthermore, model predictions of the gene dynamics are more accurate than predictions made with neural networks. This holds true even in the limit of small sample sizes. We compare predictions for various gene knockouts with published experimental results, finding substantial agreement. We also make predictions for gene dynamics under various gene network perturbations, impossible in non-mechanistic models.

systems biology↗

bioRxiv2021

Genome-Wide Covariation in SARS-CoV-2

The SARS-CoV-2 virus causing the global pandemic is a coronavirus with a genome of about 30Kbase length [Song et al., 2019]. The design of vaccines and choice of therapies depends on the structure and mutational stability of encoded proteins in the open reading frames(ORFs) of this genome. In this study, we computed, using Expectation Reflection, the genome-wide covariation of the SARS-CoV-2 genome based on an alignment of {approx} 130000 SARS-CoV-2 complete genome sequences obtained from GISAID[Shu & McCauley, 2017]. We used this covariation to compute the Direct Information between pairs of positions across the whole genome, investigating potentially important relationships within the genome, both within each encoded protein and between encoded proteins. We then computed the covariation within each clade of the virus. The covariation detected recapitulates all clade determinants and each clade exhibits distinct covarying pairs.

genomics↗

bioRxiv2020

Bioactivity assessment of natural compounds using machine learning models based on drug target similarity

Natural compounds constitute a rich resource of potential small-molecule therapeutics. While experimental access to this resource is limited due to its vast diversity and difficulties in systematic purification, computational assessment of structural similarity with known therapeutic molecules offers a scalable approach. Here, we assessed functional similarity between natural compounds and approved drugs by combining multiple chemical similarity metrics and physicochemical properties through a random forest model. As a training set, we used pair-wise similarity between 1410 drugs in terms of their shared protein targets. The resulting model featured high performance metrics (matthews correlation coefficient of 0.81, and balanced accuracy of 0.91) suggesting that it well-captured the structure-activity relation. The model was then used to predict protein targets of circa 11k natural compounds by comparing them with the drugs. This revealed therapeutic potential of several natural compounds, including those with support from previously published sources as well as those hitherto unexplored. We experimentally validated one of the predicted links activities, viz., Cox-1 inhibition by 5-methoxysalicylic acid, a molecule commonly found in tea, herbs and spices. In contrast, another natural compound, 4-isopropylbenzoic acid, which showed a higher similarity when considering the most weighted similarity metric but was not picked by the random forest model, did not inhibit Cox-1. Our results demonstrate the utility of a machine-learning approach combining multiple chemical features for uncovering protein binding potential of natural compounds.

bioinformatics↗

bioRxiv2018

A mathematical model of calcium dynamics: Obesity and mitochondria-associated ER membranes

Multiple cellular organelles tightly orchestrate intracellular calcium (Ca2+) dynamics to regulate cellular activities and maintain homeostasis. The interplay between the endoplasmic reticulum (ER), a major store of intracellular Ca2+, and mitochondria, an important source of adenosine triphosphate (ATP), has been the subject of much research, as their dysfunctionality has been linked with metabolic diseases. Interestingly, through out the cells cytosolic domain, these two organelles share common microdomains called mitochondria-associated ER membranes (MAMs), where their membranes are in close apposition. The role of MAMs is critical for intracellular Ca2+ dynamics as they provide hubs for direct Ca2+ exchange between the organelles. A recent experimental study reported correlation between obesity and MAM formation in mouse liver cells, and obesity-related cellular changes that are closely associated with the regulation of Ca2+ dynamics. We constructed a mathematical model to study the effects of MAM Ca2+ dynamics on global Ca2+ activities. Through a series of model simulations, we investigated cellular mechanisms underlying the altered Ca2+ dynamics in the cells under obesity. We found that the formation of MAMs is negatively correlated with the amplitude of cytosolic Ca2+ activities, but positively correlated with that of mitochondrial Ca2+ dynamics and the overall frequency of Ca2+ oscillations. We predict that, as the dosage of stimulus gradually increases, liver cells from obese mice will reach the state of saturated cytosolic Ca2+ concentration at a lower stimulus concentration, compared to cells from healthy mice.\n\nAuthor summaryIt is well known that intracellular Ca2+ oscillations carry encoded signals in their amplitude and frequency to regulate various cellular processes, and accumulating evidence supports the importance of the interplay between the ER and mitochondria in cellular Ca2+ homeostasis. Miscommunications between the organelles may be involved in the development of metabolic diseases. Based on a recent experimental study that spotlighted a correlation between obesity and physical interactions of the ER and mitochondria in mouse hepatic cells, we constructed a mathematical model as a probing tool that can be used to computationally investigate the effects of the cellular changes linked with obesity on global cellular Ca2+ dynamics. Our model successfully reproduced the experimental study that observed a positive correlation between the ER-mitochondrial junctions and the magnitude of mitochondrial Ca2+ responses. We postulate that hepatic cells from lean animals exhibit Ca2+ oscillations that are more robust under higher concentrations of stimulus, compared to cells from obese animals.

cell biology↗