Biology subjects

Tomic, I.

Publications and source records attributed to Tomic, I..

15 recordsLinked to original sources

bioRxiv2026

End-to-end automation of behavioural experiments with a collaborative robot

Reliable behavioural measurement is critical to neuroscience discovery, but end-to-end automation of experiments using live animals is challenging. Preparation and transfer of live animals into and out of behavioural assays require skilled, dexterous, gentle and consistent handling. This is difficult to achieve without bespoke automation systems, which resist wide adoption. We present STELA, a robotic cell featuring off-the-shelf collaborative robot (cobot) that autonomously runs the complete workflow of a Drosophila behaviour experiment, from vial to data. Designed to emulate human routine fly laboratory procedures in human workspace, it performs manipulation of fly culture vials, CO2 fly anaesthesia, vision-guided sex classification, single live fly manipulation, and loading and disassembly of a behaviour assay, all in a programmed loop. In 24-h working day, STELA completed 16 experiments (1024 individual visual learning assays in flies) compared to the human who, in an 8-hour working day, completed 6 (384 assays). Retention of flies fit for analysis was similar (93.85% versus 83.33%, respectively), but with 12-fold improvement in handling consistency by the robot. In biologically matched behaviour experiments, robotic and manual handling resulted in the same mean fly behavioural outcomes, yet the robot significantly reduced inter-individual variability in baseline and learning behaviours. STELA demonstrates that it is possible to deploy standard robotic systems to flexibly automate one of the most challenging animal experiments. It provides a low-variance, high-throughput platform for scalable experimentation, and a path to closed-loop, democratized robotic acceleration of experimental discovery.

animal behavior and cognition↗

bioRxiv2026

Learned Immune Architectures of Durable Antibody Responses Across Vaccines

Vaccination is one of the most effective public health interventions. However, vaccine efficacy varies widely among individuals, as immunity arises from complex interplay between genetic, pathogen, and immunological factors. To date, most systems vaccinology studies have remained pathogen-specific, precluding the discovery of potential shared immune architectures underlying durable antibody responses. To address this gap, we leveraged transcriptomic data from 1,032 participants receiving influenza, hepatitis B, or yellow fever vaccines to develop an interpretable machine learning framework for comparative analysis across diverse vaccine platforms. Pathogen-specific models using Blood Transcriptional Module-based feature aggregation accurately predicted high antibody responders and consistently outperformed gene-level models. Distinct predictive immune architectures identified across vaccines were further resolved for dominant hierarchical immune programs using surrogate decision trees. This approach identified the dominant decision boundaries underlying each vaccine model, highlighting leukocyte migration and Th2 differentiation in Hepatitis B, CD4+ T cells, M2 macrophages, and c-MYC signaling in Influenza, and B-cell receptor signaling with B-cell developmental pathways in Yellow Fever. Cross-pathogen concordance analyses further identified four shared transcriptional modules, suggesting partially conserved immune architectures across diverse vaccines. Together, these findings provide new insights into the immune mechanistic underpinnings of durable vaccine responses across vaccines and provide an interpretable framework for comparative systems vaccinology that may guide the rational design of next-generation vaccines. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=77 SRC="FIGDIR/small/737303v1_ufig1.gif" ALT="Figure 1"> View larger version (28K): org.highwire.dtl.DTLVardef@1646f50org.highwire.dtl.DTLVardef@c816eforg.highwire.dtl.DTLVardef@59b19eorg.highwire.dtl.DTLVardef@9c39cc_HPS_FORMAT_FIGEXP M_FIG C_FIG

immunology↗

bioRxiv2025

Innate antiviral readiness drives the expansion of protective T stem cell memory against influenza

The development of T-cell-based influenza vaccines relies on eliciting broad CD8+ T-cell immunity, wherein T stem cell-like memory (TSCM) cells serve as the ultimate long-lived reservoir for immune memory, thereby unlocking the potential for durable protection against viral drift and shift. However, the specific immunological cues that drive the robust expansion and functional preservation of this self-renewing, multipotent subset remain unknown. Here, utilizing multi-omic systems immunology in a pediatric cohort immunized with live attenuated influenza vaccine, we identified the determinants governing the expansion of influenza virus-reactive TSCM cells. We show that a pre-existing state of innate antiviral readiness, defined by a plasmacytoid dendritic cell-associated type I interferon signature, is the requisite condition for a robust TSCM expansion. Mechanistically, this baseline innate state enhances antigen priming and enforces a qualitative divergence in T-cell fate, driving responders toward a functionally poised, Th1-dominant phenotype while non-responders default to a dysfunctional, hyper-proliferative state. To determine the clinical relevance of this cellular subset, we analyzed an independent controlled human influenza challenge study. This validation revealed a critical functional division of labor in host defense: whereas pre-existing antibodies primarily mitigated symptom severity, the baseline frequency of influenza virus-reactive TSCM cells was the strongest predictor of rapid viral load clearance. These findings establish that the expansion of durable cellular memory is not stochastic but is predetermined by the innate cytokine environment, providing a predictive biomarker for patient stratification and a validated target for adjuvants designed to expand the TSCM reservoir deliberately.

immunology↗

bioRxiv2025

Intrinsic rewards guide visual resource allocation via reinforcement learning

Humans and other animals prioritise visual processing of stimuli that signal rewards. While prior research has focused on tangible incentives (e.g., money or food), the effects of intrinsic incentives - such as perceived competence - are less well understood. Across a series of visual estimation experiments, we manipulated observers subjective sense of confidence in their judgements using either deceptive trial-by-trial feedback or real discrepancies in stimulus reliability. We found that observers prioritised encoding of stimuli associated with lower uncertainty or error, benefiting performance for stimuli already estimated accurately, while further impairing performance for those estimated poorly. These reward-driven biases, while potentially adaptive, impaired overall accuracy in the present tasks by causing resource allocation to deviate from the error-minimizing strategy. To account for these findings, we supplemented a normalization model of neural resource allocation with a simple reinforcement learning rule. Intrinsic and external rewards cumulatively shaped the values assigned to different stimuli by the model, and the resulting discrepancies biased resource allocation and thereby estimation error, quantitatively matching the data. These findings reveal how intrinsic reward signals can shape resource allocation in ways that are both adaptive and counterproductive, offering a computational basis for the motivational biases underlying cognitive performance.

neuroscience↗

bioRxiv2025

Integrative Mapping of Pre-existing Immune Landscapes for Vaccine Response Prediction

Predicting individual vaccine responses remains a significant challenge due to the complexity and variability of immune processes. To address this gap, we developed immunaut, an open-source, data-driven framework implemented as an R package specifically designed for all systems vaccinologists seeking to analyze and predict immunological outcomes across diverse vaccination settings. Leveraging one of the most comprehensive live attenuated influenza vaccine (LAIV) datasets to date - 244 Gambian children enrolled in a phase 4 immunogenicity study - immunaut integrates humoral, mucosal, cellular, transcriptomic, and microbiological parameters collected before and after vaccination, providing an unprecedentedly holistic view of LAIV-induced immunity. Through advanced dimensionality reduction, clustering, and predictive modeling, immunaut identifies distinct immunophenotypic responder profiles and their underlying baseline determinants. In this study, immunaut delineated three immunophenotypes: (1) CD8 T-cell responders, marked by strong baseline mucosal immunity and extensive prior influenza virus exposure that boosts memory CD8 T-cell responses, without generating influenza virus-specific antibody responses; (2) Mucosal responders, characterized by pre-existing systemic influenza A virus immunity (specifically to H3N2) and stable epithelial integrity, leading to potent mucosal IgA expansions and subsequent seroconversion to influenza B virus; and (3) Systemic, broad influenza A virus responders, who start with relatively naive immunity and leverage greater initial viral replication to drive broad systemic antibody responses against multiple influenza A virus variants beyond those included in the LAIV vaccine. By integrating pathway-level analysis, model-derived contribution scores, and hierarchical decision rules, immunaut elucidates how distinct immunological landscapes shape each response trajectory and how key baseline features, including pre-existing immunity, mucosal preparedness, and cellular support, dictate vaccine outcomes. Collectively, these findings emphasize the power of integrative, predictive frameworks to advance precision vaccinology, and highlight immunaut as a versatile, community-available resource for optimizing immunization strategies across diverse populations and vaccine platforms. One-Sentence SummaryUsing one of the most comprehensive LAIV datasets compiled to date, immunaut, an integrative machine learning framework, identifies distinct immunophenotypic responder groups shaped by baseline immune landscapes, advancing precision vaccinology and guiding more effective, personalized immunization strategies. Graphical abstractImmunaut, an automated framework for mapping and predicting vaccine response immunotypes. Step 1 outlines the identification of vaccine response outcomes using pre- and post-vaccination data integration across immune features, including antibodies, flu-specific T-cells, and immunophenotyping at mucosal and systemic sites. Clustering methods define the vaccine response landscape, stability, and validation through t-SNE-based visualization. Step 2 leverages an automated machine learning modeling approach, to enhance the accuracy and interpretability of vaccine response predictions, enabling stratification and targeted intervention strategies for personalized vaccine immunogenicity. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=84 SRC="FIGDIR/small/634302v1_ufig1.gif" ALT="Figure 1"> View larger version (21K): org.highwire.dtl.DTLVardef@62276corg.highwire.dtl.DTLVardef@a5f25org.highwire.dtl.DTLVardef@eca34borg.highwire.dtl.DTLVardef@1d4999b_HPS_FORMAT_FIGEXP M_FIG C_FIG

immunology↗

bioRxiv2024

Genular: An Integrated Platform for Defining Cellular Identity and Function through Single-Cell Gene Expression and Multi-Domain Biological Data

Accurately defining cellular identity and function is essential for advancing immunology, understanding disease mechanisms, and developing targeted therapies. However, current bioinformatics tools are limited in their ability to integrate and analyze the vast and diverse single-cell RNA sequencing (scRNA-seq) datasets available, hindering the comprehensive capture of cellular heterogeneity and the identification of subtle genetic changes across immune states, differentiation pathways, and tissue contexts. To overcome these challenges, we introduce genular, an open-source platform that unifies gene expression data analysis across diverse cell types by integrating scRNA-seq data with extensive genomic and proteomic information from 16 databases, including NCBI Gene, Human Protein Atlas, STRING, and UniProt. genular aggregates data from more than 2,893 scRNA-seq experiments, encompassing over 74.5 million unique cells across various tissues and conditions. A key feature of genular is calculating a cell marker score for each gene, enabling the quantification of gene expression across cells to derive unique profiles specific to cell types, states, and lineages. Using genular, we differentiate T cell memory states, map differentiation profiles by tracking gene expression changes as monocytes mature into macrophages and lymphoid progenitor cells develop into T cells, and capture tissue-specific reprogramming of macrophages, revealing distinct gene expression profiles that enable specialized functions in different tissues. By integrating scRNA-seq data with multi-domain biological information and employing advanced statistical methodologies, genular provides a scalable platform that accurately defines cellular identities, functional states, and differentiation pathways. This comprehensive approach facilitates breakthroughs in immunology, gene regulation, cellular differentiation, and disease research, enabling a deeper understanding of immune cell functions and their roles in health and disease.

bioinformatics↗

bioRxiv2024

Behavioral individuality is a consequence of experience, genetics and learning

Learning and memory are essential components of our individuality. While it is established that behaviour can vary across genetically identical individuals, it remains unknown how much of this variation stems from momentary experience during learning compared to genetics and its past interactions with the environment. To address this, we measured behaviour in thousands of flies from 90 genetic backgrounds while they performed tasks in conditions that either did or did not require learning. Flies that were genetically identical, raised under the same conditions and tested simultaneously in the same environment persistently modified the extent of expressed individuality when they could learn. This learning-induced residual expression of individuality and its dynamics were subdued or absent in innate, learning-independent behaviours. We could quantify and then recreate this phenomenon in computer simulations. The emergence of in silico behavioural individuality was most consistent with the individuality of real flies once we enabled reinforced learning in simulated agents. Moreover, we showed that minor differences in initial conditions of the experiment can exacerbate the expression of individuality within a genotype in a learning-dependent manner. Our results establish that besides the classical G x E interactions shared between individuals in the past, learning from individual momentary experience further extends the expression of individuality.

animal behavior and cognition↗

bioRxiv2024

Divisive attenuation based on noisy sensorimotor predictions accounts for excess variability in self-touch

When one part of the body exerts force on another, the resulting tactile sensation is perceived as weaker than when the same force is applied by an external agent. This phenomenon has been studied using a force matching task, in which observers are first exposed to an external force on a passive finger and then instructed to reproduce the sensation by directly pressing on the passive finger with a finger of the other hand: healthy participants consistently exceed the original force level. However, this exaggeration of the target force is not observed if the observer generates the matching force indirectly, by adjusting a joystick or slider that controls the force output of a motor. Here we present the first detailed computational account of the processes leading to exaggeration of target forces in the force matching task, incorporating attenuation of sensory signals based on motor predictions. The model elucidates previously unappreciated contributions of multiple sources of noise, including memory noise, in determining matching force output, and shows that quantifying attenuation as the discrepancy between direct and indirect self-generated forces isolates its predictive component. Our computational account makes the prediction that attenuated sensations will display greater trial-to-trial variability than unattenuated ones, because they incorporate additional noise from motor prediction. Quantitative model fitting of new and existing force matching data confirmed the prediction of excess variability in self-generated forces and provided evidence for a divisive rather than subtractive mechanism of attenuation, while highlighting its predictive nature. NEW & NOTEWORTHYWe formulate a detailed computational account of sensory attenuation in force matching tasks that disambiguates contributions of perceptual, memory and prediction noise in order to isolate a pure measure of attenuation strength. Analysis of data from nearly 500 participants shows that attenuated sensations display increased trial-to-trial variability, consistent with incorporating additional noise inherent to motor prediction. These results support a divisive - rather than subtractive - reduction in the sensation of self-generated forces based on predicted reafference.

neuroscience↗

bioRxiv2023

Modulation of Alzheimer's Disease Brain Pathology in Mice by Gut Bacterial Deletion: The Role of Il-17a and Microglial MyD88

Gut bacteria regulate brain pathology of Alzheimers disease (AD) patients and animal models; however, the underlying mechanism remains unclear. In this study, 3-month-old APP-transgenic female mice with and without knock-out of Il-17a gene, or haploinsufficiency of MyD88 in microglia were treated with antibiotics-supplemented or normal drinking water for 2 months. Antibiotic treatment eradicated gut bacteria, particularly in the phyla Bacteroidetes and Firmicutes, and reduced Il-17a-expressing CD4-positive T lymphocytes. Deletion of gut bacteria inhibited inflammatory activation in the brain and microglia, and reduced cerebral A{beta} levels in APP-transgenic mice, which was abolished by deficiency of Il-17a or haploinsufficiency of MyD88 in microglia. As possible mechanisms regulating A{beta} pathology, deletion of gut bacteria inhibited {beta}-secretase activity and increased the expression of Abcb1 and Lrp1 in the brain or at the blood-brain barrier, which were also reversed by the absence of Il-17a. Interestingly, a crossbreeding experiment between APP-transgenic mice and Il-17a knockout mice further showed that deficiency of Il-17a had already increased Abcb1 and Lrp1 expression at the blood-brain barrier. Thus, deletion of gut bacteria attenuates inflammatory activation and amyloid pathology in APP-transgenic mice via Il-17a and microglial MyD88-involved signalling pathways. Our study contributes to a better understanding of the gut-brain axis in AD pathophysiology.

neuroscience↗

bioRxiv2023

A dynamic neural resource model bridges sensory and working memory

Probing memory of a complex visual image within a few hundred milliseconds after its disappearance reveals significantly greater fidelity of recall than if the probe is delayed by as little as a second. Classically interpreted, the former taps into a detailed but rapidly decaying visual sensory or "iconic" memory (IM), while the latter relies on capacity-limited but comparatively stable visual working memory (VWM). While iconic decay and VWM capacity have been extensively studied independently, currently no single framework quantitatively accounts for the dynamics of memory fidelity over these timescales. Here we extend a stationary neural population model of VWM with a temporal dimension, incorporating rapid sensory-driven accumulation of activity encoding each visual feature in memory, and a slower accumulation of internal error that causes memorized features to randomly drift over time. Instead of facilitating read-out from an independent sensory store, an early cue benefits recall by lifting the effective limit on VWM signal strength imposed when multiple items compete for representation, allowing memory for the cued item to be supplemented with information from the decaying sensory trace. Empirical measurements of human recall dynamics validate these predictions while excluding alternative model architectures. A key conclusion is that differences in capacity classically thought to distinguish IM and VWM are in fact contingent upon a single resource-limited WM store.

neuroscience↗

bioRxiv2022

Perceptual similarity judgments do not predict the distribution of errors in working memory

Population coding models provide a quantitative account of visual working memory (VWM) retrieval errors with a plausible link to the response characteristics of sensory neurons. Recent work has provided an important new perspective linking population coding to variables of signal detection, including d-prime, and put forward a new hypothesis: that the distribution of recall errors on, e.g., a color wheel, is a consequence of the psychological similarity between points in that stimulus space, such that the exponential-like psychophysical distance scaling function can fulfil the role of population tuning and obviate the need to fit a tuning width parameter to recall data. Using four different visual feature spaces, we measured psychophysical similarity and memory errors in the same participants. Our results revealed strong evidence for a common source of variability affecting similarity judgments and recall estimates, but did not support any consistent relationship between psychophysical similarity functions and VWM errors. At the group level, the responsiveness functions obtained from the psychophysical similarity task diverged strongly from those that provided the best fit to working memory errors. At the individual level, we found convincing evidence against an association between observed and best-fitting similarity functions. Finally, our results show that the newly proposed exponential-like responsiveness function has in general no advantage over the canonical von Mises (circular normal) function assumed by previous population coding models.

neuroscience↗

bioRxiv2021

Deficiency of p38α-MAPK in myeloid cells ameliorates symptoms and pathology of APP-transgenic Alzheimer's disease mice

Microglial activation is a hall marker of Alzheimers disease (AD); its pathogenic role and regulating mechanisms are unclear. p38-MAPK, a stress-responding kinase, is activated in AD brain in early disease stages. In APP-transgenic mice, we deleted p38-MAPK in whole myeloid cells from birth or specifically in microglia from 9 months, and analysed AD pathology at the age of 4, 9 and 12 months. In both experimental settings, p38-MAPK deficiency decreased cerebral A{beta} and improved cognitive function of AD mice; however, p38-MAPK-deficient myeloid cells were more effective than p38-MAPK-deficient microglia in preventing AD pathogenesis. Deficiency of p38-MAPK in myeloid cells inhibited the inflammatory activation of individual microglia by 4 months, but enhanced it by 9 months. Inflammatory activation was essential for p38-MAPK deficiency to promote microglial internalization of A{beta}. Interestingly, p38-MAPK deficiency in peripheral myeloid cells reduced il-17a transcription in CD4-positive spleen cells. By cross-breeding APP-transgenic mice and IL-17a knockout mice, we further observed that IL-17a deficiency activated microglia and decreased A{beta} deposits in AD mouse brain. Thus, p38-MAPK deficiency in myeloid cells prevents AD pathogenesis, perhaps through reducing IL-17a-expressing T lymphocytes, and promoting A{beta} clearance in the brain. Our study supports p38-MAPK as a novel target for AD therapy.

neuroscience↗

bioRxiv2020

SIMON: open-source knowledge discovery platform

Data analysis and knowledge discovery has become more and more important in biology and medicine with the increasing complexity of the biological datasets, but necessarily sophisticated programming skills and in-depth understanding of algorithms needed pose barriers to most biologists and clinicians to perform such research. We have developed a modular open-source software SIMON to facilitate the application of 180+ state-of-the-art machine learning algorithms to high-dimensional biomedical data. With an easy to use graphical user interface, standardized pipelines, automated approach for machine learning and other statistical analysis methods, SIMON helps to identify optimal algorithms and provides a resource that empowers non-technical and technical researchers to identify crucial patterns in biomedical data.

systems biology↗

bioRxiv2019

The FluPRINT dataset: A multidimensional analysis of the influenza vaccine imprint on the immune system

Recent advances in machine learning have allowed identification of molecular and cellular factors that underly successful antibody responses to influenza vaccines. Results of these studies have revealed the high level of complexity necessary to establish influenza immunity, and many different cellular and molecular components involved. However, identified correlates of protection fail to account for the majority of vaccinated cases across ages, cohorts, and influenza seasons. Major challenges arise from small sample sizes and from analysis of only one aspect of the biology such by using transcriptome data. The objective of the current study is to create a unified database, entitled FluPRINT, to enable a large-scale study exploring novel cellular and molecular underpinnings of successful immunity to influenza vaccines. Over 3,000 parameters were considered, including serological responses to influenza strains, serum cytokines, cell subset phenotypes, and cytokine stimulations. FluPRINT, thus facilitates application of machine learning algorithms for data mining. The data are publicly available and represent a resource to uncover new markers and mechanisms that drive successful influenza vaccination.

systems biology↗

bioRxiv2019

SIMON, an automated machine learning system reveals immune signatures of influenza vaccine responses

Machine learning holds considerable promise for understanding complex biological processes such as vaccine responses. Capturing interindividual variability is essential to increase the statistical power necessary for building more accurate predictive models. However, available approaches have difficulty coping with incomplete datasets which is often the case when combining studies. Additionally, there are hundreds of algorithms available and no simple way to find the optimal one. Here, we developed Sequential Iterative Modelling "OverNight" or SIMON, an automated machine learning system that compares results from 128 different algorithms and is particularly suitable for datasets containing many missing values. We applied SIMON to data from five clinical studies of seasonal influenza vaccination. The results reveal previously unrecognized CD4+ and CD8+ T cell subsets strongly associated with a robust antibody response to influenza antigens. These results demonstrate that SIMON can greatly speed up the choice of analysis modalities. Hence, it is a highly useful approach for data-driven hypothesis generation from disparate clinical datasets. Our strategy could be used to gain biological insight from ever-expanding heterogeneous datasets that are publicly available.

immunology↗