Frontiers in Systems Biology

Rule-based languages have proven effective for modeling systems of interacting structured entities as typically encountered in chemistry and molecular biology. We present PyKappa, a rule-based modeling package written in Python whose interpreted nature enables interactive simulation and analysis, including by agentic AI. The package seeks to broaden the base of developers by utilizing a widely known programming language and serves as an easy-to-deploy teaching tool. Using PyKappa, we conduct a case study of phase separation.

Oscillatory and saturating dynamics in regulatory gene networks govern processes such as cell cycle timing, circadian rhythms, and inflammatory signaling, and sparse regression methods such as SINDy offer an interpretable method for recovering their governing equations from time-series data. However, these networks are typically governed by Hill-type nonlinearities that become nearly collinear with polynomial candidates over most bounds of an oscillating curve, and placing them in a shared library produces ill-conditioned regressions and unreliable coefficient estimates. Standard approaches such as ridge regularization, library normalization, and orthogonalization can marginally improve conditioning but cannot remove the structural multicollinearity of these systems, in turn limiting our ability to produce accurate parameter estimates. To address this issue, we propose the Regulation-Informed Sparse Identification of Nonlinear Dynamics (RI-SINDy) method, which removes known regulatory terms from the regression and recovers their coefficients directly from an asymptotic drift-balance condition. Because the time averaged derivative of a bounded trajectory approaches zero, the coefficients of regulatory nonlinearities can be computed independently by constraining them to offset the mean contribution of the remaining terms. We demonstrate that this method effectively reduces library collinearity and improves conditioning numbers by 3-45x, substantially improving coefficient recovery and regression accuracy. We test the RI-SINDy method on experimental Hes1 data and on synthetic Goodwin and NF-{kappa}B models. On Hes1, it achieves the lowest trajectory error among the compared methods (Traditional SINDy, SR3, and Nullcline Reconstruction). On the synthetic systems, it recovers the correct sparse structure with lower trajectory error than SR3 and Nullcline Reconstruction, and in a Goodwin noise sweep it recovers all active terms at noise levels up to 15%, where Traditional SINDy and E-SINDy fail to. In conclusion, the RI-SINDy method provides a robust framework to overcome the issue of multicollinearity for system identification in biological systems.

Tsetse flies are strictly found is sub-Saharan Africa where they are responsible for the transmission and maintenance of African trypanosomiases in humans (HAT) and animals (AAT). Vector control has been recognized as an essential tool to fight against these diseases. Nevertheless, it requires the best possible knowledge of the biology of the targeted population. Population genetics tools can prove very useful to obtain such information but require the use of polymorphic and reliable genetic markers. In this paper we present the development of microsatellite markers using a new high-throughput sequencing based technology (SSRseq). We applied it on two species of tsetse flies from three HAT foci and obtained more accurate results as compared to microsatellite loci developed with classic methods. We could indeed use 9 to 14 SSRseq loci without the several problems generally met with classic microsatellites, as all were located in autosomes, without short allele dominance or stuttering and very few null alleles. SSRseq loci appeared much more polymorphic in tsetse flies as compared to other species (fungi, trees, bees, or fishes), which suggested much higher effective population sizes, much higher mutation rates of the genome or both, suggesting high capacities for evolutionary adaptation. With the 9-14 loci kept, we confirmed the propensity of these flies to move almost freely in the whole zones investigated, and also highlighted the possible evolutionary response of one of the unkept loci regarding vector control devices used in these HAT foci, which will require further studies. We suggest for further population structure studies to use only loci with less than 1% missing data, which proved being a good and fast selection strategy to get the most reliable results.

The modeling of aroma release in food science is a complex, multidisciplinary field integrating mass transfer theory and principles of physical chemistry. While mechanistic models have long been used to understand volatile release, their practical adoption has been hindered because existing implementations are often private, "in-house" tools that lack reproducibility and extensibility. This paper introduces digimouth, an open-source Python framework designed to provide a unified, modular, and reproducible environment for simulating and fitting aroma release kinetics. A key innovation of digimouth is its handling of events and stages. Events represent discrete, instantaneous changes in the system (such as opening a bottle or swallowing), while stages allow for multi-phase simulations where experimental conditions or physical dynamics change over time. To demonstrate the tool's utility, the authors implemented established in vitro convection models and performed sensitivity analyses on parameters such as air-water partition coefficients and mass transfer coefficients. The package also includes an optimization module based on the least-squares method to fit parameters to experimental data. Validation against real-time data from Proton Transfer Reaction-Time of Flight- Mass Spectrometry (PTR-ToF-MS) shows a satisfactory fit, with an average error of approximately 2%. Ultimately, digimouth facilitates robust model development and hypothesis testing, offering the food science community a collaborative tool to better understand the mechanisms of flavor release.

The perception of flavor during food consumption is a complex process governed by the dynamic release of aroma compounds, which is influenced by the interaction between the food matrix and human physiology. While mechanistic models have been developed to simulate these phenomena, their practical application has often been limited by the lack of flexible software tools capable of integrating diverse food structures and real-time physiological data. This study addresses this gap by incorporating in vivo mechanistic models for both liquid foods and melting solid matrices in a modular Python-based framework: digimouth. The modeling approach treats the oral, pharyngeal, and nasal cavities as a series of interconnected reactors with mass transfer governed by ordinary differential equations (ODEs). These models explicitly integrate key physiological parameters, including salivary flow, bidirectional respiratory airflow, and discrete events such as mastication and swallowing. For solid foods, the framework incorporates additional mechanisms for matrix dissolution, temperature-dependent melting, and dynamic surface area change. The models were validated using experimental data from 30 subjects who consumed aroma solutions and flavored gelatin gummy discs while monitored by PTR-MS (Proton Transfer Reaction - Mass Spectrometry). The simulations successfully reproduced the main kinetic features and release profiles observed experimentally for both types of matrices. Despite some limitations related to the self-reporting of physiological events and the non-inclusion of initial orthonasal mass transfer in the models, the developed software package proved to be a robust and evolutive tool. It offers a foundation for exploring inter-individual variability and the impact of complex food structures on sensory perception.

Background. Blood-based biological age clocks predict mortality well, but they are statistical composites. They do not say which physiological systems make a person biologically older, and they cannot say how an intervention would move the clock or whether moving it would change risk. Methods. We built IQN-BIOAGE-01, a quantitative systems pharmacology (QSP) model of mortality-referenced biological age. A systemic senescent load follows saturating-removal stochastic dynamics and drives eight latent organ/hallmark axes, which are observed through 13 routine blood, blood pressure and body size measurements. Death is the first passage of the load over a threshold lowered by poor organ reserve, and biological age (BAmech) is the age at which the reference population has the same 10year risk. The model was calibrated on NHANES III (14,049 adults; 2,736 deaths in 20 years) and tested on a locked NHANES IV 1999 2010 holdout (24,650 adults; 1,867 deaths). Two later versions were accepted or rejected by rules written before their validation results. We confronted the inflammation axis with three anti-inflammatory outcome trials, audited common implementations of the comparator clock, and locked a registration-ready external validation. Results. On the hold-out, Harrell's C was 0.842 for BAmech, 0.842 for PhenoAge and 0.823 for chronological age. The hazard ratio per SD of the biological age gap was 1.55 (1.48 1.62) for BAemech and 1.45 (1.40 1.50) for PhenoAge. Risk ordering transferred across eras (calibration slope 1.01 1.03), but absolute 10year risk was overpredicted (10.4% against 7.4% observed); life-table recalibration moved observed-to-expected from 0.72 to 0.81. The inflammation axis carried most of the gap. A joint re-estimate did not converge and a converged, biological-variation-anchored version narrowly failed its pre-written identifiability criterion, so v1.0 remains the locked primary. The outcome evidence available so far, chiefly topline results from one anti-inflammatory trial, suggested that only a small fraction of a pharmacological CRP fall may carry the risk implied by the model's cross-sectional structure (posterior median pass-through 0.22). Common software implementations of PhenoAge differed from the published formula by 0.7 2.0 years. Conclusions. A mechanistic model can match a leading statistical blood clock in discrimination while adding an organ-level decomposition, explicit statements of what the data can and cannot identify, and a structure in which interventions can be simulated. Because inflammation dominates blood-based biological age while lowering CRP with drugs may carry little of the implied risk, a change in a blood clock should not be read as a change in mortality risk without outcome evidence. Validation is Tier 1 (internal and temporal, within one survey programme); the external test is locked and awaiting data access.

Bioinformatics software plays a central role in biological research, yet many academic projects lack robust testing and continuous integration (CI) practices. This absence can compromise reproducibility, accessibility, and long-term maintainability. In this study, we conduct an empirical analysis of GitHub repositories spanning bioinformatics and other scientific software domains to examine the relationship between CI, software testing, and software quality. Repository discovery was performed using domain-specific search terms via GitHub's Application Programming Interface (API). This was followed by a layered data extraction pipeline, which included a Large Language Model (LLM) validation step, that connected repositories to published works, institutions, authors, funding, and citation counts. By integrating repository activity, community engagement, and backlog health into a composite Repository Quality Index (RQI), we find that the presence of CI and software tests is significantly associated with higher repository quality. Additionally, repositories with higher academic citation counts tend to cluster in higher RQI tiers, suggesting that software quality, which is driven by proper testing and integration, is associated with academic impact and funding status. These results demonstrate that adopting industry-standard development practices, such as CI and software testing, is a tractable path toward more sustainable and impactful academic software and funding opportunities.

When an ontology-based biomedical knowledge graph (a KG, containing information about genes, pathways, phenotypes, and molecular interactions) is analyzed using a computational reasoning system, the system can return unwanted generic concepts. Such generic concepts, in the context of a query about potential drug treatments for a disease, might include 'pharmaceutical preparations' or 'antibiotics'. Answers can be filtered based on hand-curated block-lists or ontology Information Content (IC), but to our knowledge, (1) neither filtering method has been validated against expert judgments of generality, and (2) IC is not available for the majority of nodes in the representative KG used in this work. In this study, three domain experts rated representative lists of KG concepts for generality. We built a semantic precision score (SPS) from a concept's name and description alone, using a biomedical language model (BioLORD). For the purpose of filtering generic answers from biomedical query results, we compared (using queries for 200 diseases) the SPS against a curated concept block-list and two IC thresholds. IC agreed moderately with the experts (Spearman's {rho} = 0.59), but correlations were negligible or negative within IC bands. The SPS was well-correlated with the experts ({rho} = 0.76) globally and within SPS bands, and remained informative on concepts without IC ({rho} = 0.59). Hard filtering of query results using the SPS removed 1,502 of 33,161 query answers, nearly three times the block-list's 513, while removing fewer of the query-specific DrugCentral indications. Hard filtering by IC removed either few answers or many approved drugs. We additionally compared the filtering methods' performance on an independent evaluation set of queries for 200 diseases. The SPS filter removed 5,135 of 91,203 answers compared with the block-list's 1,419 and lost none of 1,296 indications compared with 17. Further, SPS filtering removed approved drugs no more often from the half of DrugCentral not used in training than from the half used in training. Semantic meaning captures expert judgments of specificity more faithfully than ontology position, and semantic scoring can replace IC and manually maintained block-lists for filtering drug-repurposing results.

Moment-closure methods provide a widely used approach for approximating low-order statistics of stochastic biochemical reaction networks, but the resulting closed equations may exhibit physically meaningless dynamics. Here, we systematically investigate the stability of second-order moment-closure approximations for a bursty gene expression model with nonlinear protein degradation. We consider the Gaussian, log-normal, and negative-binomial closures. By analyzing the resulting dynamical systems, we show that the Gaussian closure fails the proposed stability criterion for all parameter values, whereas both the log-normal and negative-binomial closures possess a unique physically accessible fixed point and are stable if and only if kB(1 + 2B) [≥] 1, where B is the mean burst size and k is the dimensionless protein synthesis rate. Moreover, this condition is automatically satisfied when the steady-state mean protein copy number is at least one, suggesting that it is mild in biologically relevant regimes. Numerical comparisons over a broad parameter range further show that the negative-binomial closure is the most accurate, while the Gaussian closure is the least accurate. We further develop higher-order negative-binomial moment-closure schemes and demonstrate that increasing the closure order substantially improves predictive accuracy. These results highlight the importance of considering dynamical stability, in addition to numerical accuracy, when assessing moment-closure approximations.

Obesity is a major health concern worldwide and is caused by an excessive fat accumulation in the body due to adipogenesis. Recent studies suggest that excessive stress or disruption of circadian rhythms can enhance adipogenic commitment and lead to obesity. Moreover, this adipogenic commitment occurs at a specific time of day, highlighting a correlation with the circadian rhythm. However, the dynamical nature of this kind of stress-induced adipogenic differentiation remains an unresolved puzzle. Herein, we proposed a mathematical model centred on the key transcription factor PPAR{gamma}, associated with adipogenic differentiation, and demonstrated that adipogenic commitment is a bistable, irreversible differentiation process in which slow positive feedback regulation of PPAR{gamma} primarily controls the differentiation commitment threshold. The model reproduced the experimental observations regarding the extent of adipogenesis induced by constant and pulsatile stress input signals and explained the correlation between adipogenic differentiation time and circadian phase. It further predicts how to control the adipogenic commitment even under stressful conditions. Overall, our study elucidates how stress signalling dynamically governs adipogenesis.

Agentic artificial intelligence (AAI) is increasingly used by biomedical scientists, where it has substantially lowered the difficulty of integrating computational, statistical and data-science approaches into day-to-day research activities, assisting with project design, analysis and manuscript preparation. Significant barriers remain, however, which are unlikely to be removed solely by releasing more capable AI models. Meaningful use still demands a technical fluency that many biomedical researchers lack: a PI and their lab members must learn how to use the AAI system, establish standard operating procedures (SOPs) for managing, sharing and curating data, and enforce rules that address data privacy and security. Existing commercial AAI systems, which were designed primarily for software engineering purposes, need to be significantly re-purposed for biomedical-related data science. We present Murmurent, shared software that sits beneath the agentic AI and provides the skills, procedures and protections to address the concerns above, removing the need for a lab to develop them independently. This includes (1) infrastructure to support projects and so-called choreographies} involving multiple group members; (2) a set of specialized agents each dedicated to typical biomedical data science tasks including software and statistical development, data visualization, literature search, equity, diversity, inclusion, and decolonization (EDID) review, and others; (3) a tiered memory which can retain important research decisions, intermediate, derived outputs and sensitive data, and which can exploit this information to automatically build better contexts in AAI sessions; (4) traceability records which are used by the AAI session to better plan and execute data analyses and software builds by, for example, helping to avoid repeating decisions that lead to dead-ends; (5) enforcement of SOPs for data maintenance and data governance across all lab members; and (6) multi-user capacity to allow groups to interact and collaborate. We use the system to identify putative inhibitors of Peptidyl-prolyl cis-trans Isomerase NIMA-interacting 1 (Pin1), a protein with a shallow catalytic site making it difficult to target. We describe the construction of several approaches to identify Pin1 inhibitors and the results they yield. Murmurent is open source.

In recent years, the issue of reproducibility in scientific studies has been at the forefront of scientific discussion. Many of these issues arise due to poor employment of statistical methods, making it crucial to address these knowledge gaps for frequently used methods. Elastic net regression (ENET) is one such algorithm used frequently in biology and public health due to its ability to select important predictors (also known as "features") from large data sets. However, ENET struggles with consistent feature selection and hyperparameter choice in these applications due to low observation counts and noisy data. This limitation hinders ENET's effectiveness in biological research due to low model reproducibility and predictor interpretability. We propose the Instability Quotient Criterion (IQC) as a metric to quantify the consistency of selected features for feature selection models such as ENET, with a specific emphasis on biological applications. IQC leverages ensemble modeling and Shannon's entropy in order to determine the stability of selected features derived from ENET models. We demonstrate IQC's robustness via simulation studies, showing that IQC is capable of detecting significant decreases in model stability with lower observation counts, higher dimensionality, and higher proportions of noisy signal. We additionally utilize simulation to provide a naive threshold for low versus high stability. Finally, we applied IQC to a case study to illustrate how our method can be integrated into a larger analysis pipeline when looking at microarray gene expression data. IQC indicated stability issues for differentially-expressed genes when classifying acute leukemia subtypes in spite of high classification success. Lastly, we discuss how model instability impedes interpretation of biologically significant genes for discerning cancer subtypes. Overall, we present IQC as a robust evaluation metric that can easily be incorporated into high-dimensional biological analyses.

Background: Chronic low-grade inflammation and gut microbiota dysbiosis contribute to metabolic syndrome (MS). Huanglian Wendan Decoction (HLWDD) is a defined multi-herbal formulation used in Chinese medicine, but its effects on gut microbiota-associated metabolism and NLRP3 inflammasome signaling remain unclear. Objective: This study evaluated the pharmacological effects and candidate absorbed bioactive constituents of HLWDD in experimental MS. Methods: A high-fat, high-sugar diet-induced rat model of MS was assessed using metabolic, biochemical, histological, molecular, and multi-omics endpoints. Gut metagenomics, untargeted metabolomics of liver and colonic contents, liver transcriptomics, and UPLC-HRMS/MS analysis of serum-absorbed constituents were integrated with network pharmacology and molecular docking. Baicalin was further evaluated in PA-BSA-treated HepG2 cells. Results: HLWDD ameliorated obesity, dyslipidemia, hepatic steatosis, and systemic inflammation and attenuated NLRP3 inflammasome-related signaling. Multi-omics analysis identified nine shared differential metabolites and a strong positive association between Faecalibacterium and butyrate. Thirty-two serum-absorbed constituents were tentatively identified. Baicalin showed favorable predicted interactions with NLRP3 (-5.82 kcal/mol) and CASP1 (-5.41 kcal/mol) and concentration-dependently suppressed NLRP3 inflammasome-related signaling in HepG2 cells; the highest concentration increased intracellular butyrate. Conclusion: HLWDD ameliorated metabolic and inflammatory abnormalities in experimental MS in association with restoration of gut-liver metabolic homeostasis and attenuation of NLRP3 inflammasome signaling. Baicalin emerged as a candidate absorbed bioactive constituent.

A compound's toxicity can be gauged from the toxicities of the compounds it resembles. We showed that in 2003 for rat oral LD50: the compounds in a reference set of 13,645 that most resemble the query form a consortium and vote on its toxicity. This work returns to that method twenty-three years later, on public data, because the reference set used in 2003 is no longer public. Rebuilt on 11,989 curated rat oral LD50 values, the method answers for 67.0% of the held-out compounds carrying an exact value, and 92.1% of those answers land within one log unit; in 2003 the figures were 94% within one log unit on 57.3% of compounds. We then ran it over 28 further endpoints, each recording whether a compound showed a particular toxicity, from cytochrome inhibition to mutagenicity to clinical toxicity: 149,224 measurements on 34,162 compounds. The method answers only when a compound has close enough neighbors, and on those it correctly identifies a median 72.9% of the compounds that showed the toxicity and 96.5% of those that did not. Separately we asked whether a reverse screen carries information about these endpoints beyond molecular size and lipophilicity. We search a compound against a published index of 27,910 ligands solved inside proteins and count the protein families those sites belong to. A compound whose matches reach no protein family inhibits CYP3A4 2.5% of the time; one whose matches span sixteen or more families inhibits 51.2% of the time. Added to heavy atom count, molecular weight and cLogP, the screen raises the mean area under the ROC curve from 0.72 to 0.76 across the 28 endpoints. The OpenADMET consortium has since measured CYP3A4 inhibition on 2,335 compounds, and the same pattern holds. Compounds that resemble known binders of sixteen or more protein families inhibit CYP3A4 24.6% of the time; compounds whose matches reach no protein family inhibit it 7.3% of the time. The more protein families a compound resembles binders of, the more likely it is to inhibit CYP3A4. This association is strong enough to act on before anything is made. Scoring a library on how far its compounds reach across protein families, together with their size and lipophilicity, and setting aside the riskiest fifth removes two thirds of the androgen receptor binders and about half of the CYP3A4 inhibitors. Scoring needs only a two-dimensional structure, with no docking and no conformer generation at query time. Interactive scoring, the data and the pool are at toxpred.ai. Keywords: toxicity prediction; structural similarity; nearest neighbors; reverse screening; target breadth; promiscuity; LD50; cytochrome P450; virtual library triage

Many studies using mouse models of human brain disorders rely on extended recordings of brain electrical activity through electroencephalogram, or EEG. The features that researchers measure in these recordings, such as seizures and epileptiform activity, are deeply influenced by whether the animal is awake, in non-rapid eye movement (NREM) sleep, or in rapid eye movement (REM) sleep. Because total sleep duration varies naturally across animals, rigorous experimental comparisons require classifying the sleep state of every few seconds of a recording. Manual annotation is highly labor-intensive, often requiring hours per subject. While automated scoring tools accelerate this process, they are typically trained exclusively on healthy cohorts and frequently fail when applied to disease models that exhibit atypical or fragmented sleep architecture. Somnus is an open-source Python package designed to automate rodent sleep staging using EEG, muscle activity (EMG), and optional video-tracking data. It includes a robust base model trained on diverse recordings from six independent laboratories, enabling out-of-the-box classification across various equipment setups without requiring prior manual labeling. A central feature of Somnus is its capacity for rapid adaptation to pathological sleep structure. The software provides a graphical interface that directs researchers to the specific epochs where the base model is least certain. Users can manually correct these targeted epochs in a built-in viewer by comparing the model's classification to the raw EEG signal and video, and then use their corrections to fine-tune the model. This workflow empowers researchers to generate accurate custom scorers for complex disease models or novel treatments with just a few minutes of manual correction, bypassing the need to hand-score entire experimental cohorts or develop their own models from scratch.

Genotype imputation offers a biological application for optimization methods that may eventually run on larger quantum computers. We developed QuImputer to encode allele-frequency priors and signed linkage disequilibrium (LD) in a reduced Ising/QUBO objective, with only missing genotypes represented as optimization variables. The same objective can be passed to exhaustive classical search, quantum approximate optimization algorithm simulation, or quantum hardware. We evaluated the formulation on 100 rice accessions across 12 chromosomes, estimating LD and allele frequency from the same post-masking observed matrix supplied to Beagle 5.4. Evaluation was restricted to known truth at sites with exact dosage coding and an alternate allele that was not the major allele, with MAF below 0.50, yielding 1.88 billion scored genotypes. Beagle achieved higher rare-ALT sensitivity (2.52% versus 0.54%), F1 (0.047 versus 0.011), Matthews correlation coefficient (0.088 versus 0.044), and pooled hard-call dosage R-squared (0.231 versus 0.014). QuImputer produced fewer false ALT calls, with Type I error of 0.038% versus 0.223%. Analyses of a chromosome-1 subset showed limited local LD support for many masked rare-ALT carriers. In five small IBM Fez validation problems, the selected minimum-energy outputs matched unique all-reference exact optima; the pooled frequency of those optima was 1.21-2.63%. Further IBM Fez executions covered eleven sparse instances with 12-97 variables and a separate 120-variable instance, retaining all measurements. Selected outputs recovered the exact optima in eight instances through 49 variables. Larger instances remained below the all-reference baseline; noiseless comparisons distinguished finite-shot sampling from deviations in hardware output. QuImputer provides an LD-informed formulation and a route to quantum-hardware execution, together with an empirical assessment of the limitations of the evaluated configuration. These results establish a basis for studying richer information models and alternative classical or quantum solvers.

Several tools are available to detect differential gene expression. The predominant approach of different statistical methods is to consider changes in gene expression independent from their genomic context. Previously, DELocal was introduced that takes into consideration gene expression dynamics in the genomic neighbourhood. Whereas DELocal was shown to identify developmentally regulated genes in developing mouse tooth, its performance in other biological conditions remains undetermined. To benchmark DElocal together with other methods, we develop another tool 'broadSeq'. It provides easy to use interface for methods like DELocal , deseq2, limma, edger, EBSeq and many more. Here we report benchmarking results on different data sets that comprise wild type developing mouse organs from different embryonic time points, and also human cancer data. The examined datasets are from the heart, kidney, liver, brain, forebrain, and hindbrain. We found that most methods produce relatively comparable lists of differentially expressed genes. Nevertheless, DELocal identifies partially distinct sets of genes compared to the other methods. Evaluating the identified genes using gene and disease ontology terms show that the uniquely detected genes by DELocal are biologically relevant and also highly expressed compared to the genes identified by the other methods. Because consistently highly expressed genes show relatively subtle fold changes, the neighbourhood concept of DELocal appears to correct for the difficulty in detecting these differential expression of highly expressed genes. Finally, we provide the tools as combined bioconductor packages, allowing researchers to explore and decide the best methods for their data and questions.

Scientific discovery seeks mechanisms that explain observations and predict beyond the measurements that produced them. Whereas large language models (LLMs) encode knowledge implicitly, mechanistic world models organise it as a parsimonious set of explicit, modular, reusable mechanisms whose predictions can be scored against data. In biology, where measurements are sparse and noisy, mechanistic world modelling must discover latent states and governing equations jointly, yet the prior knowledge that could constrain this search is largely unstructured. We introduce gemot, an agentic framework for mechanistic world modelling, and evaluate it on 18 published biological problems spanning molecular biology, epidemiology, and immune-cell differentiation, with up to 1,755 training measurements, 65 observables, 170 experimental conditions, and 3 data modalities per problem. gemot is auditable by construction: a semantic layer records each hypothesis, and a Model Context Protocol layer separates hypothesis formulation from numerical evaluation, so hypothesis scoring cannot be fabricated. On every problem, autonomously constructed models match or exceed the reference models in fit and parsimony, and generalise where held-outs exist. We demonstrate that gemot can formulate novel, biologically plausible mechanistic hypotheses. This work will enable decoding of interventional biological data into competing mechanistic hypotheses and design experiments that distinguish between them, and may serve as a blueprint for mechanistic world modelling in other domains.

To reduce health risks for human space exploration, it is important to model the effects of deep-space radiation on biological systems. The budding yeast Saccharomyces cerevisiae is a common model organism in space radiobiology, but little is known about how radiation damage to individual cells translates into the population-level effects that experiments measure. The Agent-Based Model for Microbial Populations Exposed to Radiation (AMMPER) is a spatially explicit agent-based model that simulates the effects of ionizing radiation on S. cerevisiae cells and produces population-level results. Its goal is to improve experiment design and data interpretation. AMMPER-1 demonstrated that agent-based simulations can qualitatively reproduce the effects of proton radiation on the growth rate of wild-type and DNA-repair mutant yeast. Here, we present AMMPER-2, which addresses limitations of the first model and expands its capabilities. AMMPER-2 incorporates the dynamics of the redox dye alamarBlue, a data type commonly generated in spaceflight microbiology experiments, and models the fate of reactive oxygen species in more detail. It also adds an improved graphical interface and runs faster. AMMPER-2 remains easy to execute in Python on a standard personal laptop, providing space biologists a tool to explore experimental conditions, generate hypotheses, and interpret data from microbial space radiation experiments.

Dysfunctions in transcription cause severe human diseases. In eukaryotes, transcription initiation by RNA polymerase (Pol) II requires the formation of the pre-initiation complex (PIC) composed of general transcription factors (GTFs), Pol II, and an essential multiprotein coactivator, Mediator. In vitro experiments with GTFs and Pol II have depicted a linear sequence of PIC assembly. However, PIC assembly paths in vivo at the genomic scale and Mediator's role in these paths remain poorly understood. One strategy characterizes ChIP-seq occupancy alterations in Mediator mutants, but analysis of these data still lacks a formalized interpretation framework. In this study, we address this need by developing a quantitative framework that scores candidate assembly paths according to their ability to explain these alterations. This framework combines a model of PIC assembly based on a system of differential equations with parameter optimization based on a penalized criterion. We applied this new framework to datasets from two yeast Mediator mutants, in the Med17 and Med10 subunits. Our results show that the best-supported PIC assembly paths differ from those inferred in vitro. In particular, we show that it is important to distinguish the core TFIIH module from the TFIIK kinase module, and that similar paths starting with Mediator or TBP receive different levels of support. Furthermore, although multiple paths fit the data equally well, they differ markedly for specific gene groups. Taken together, the results show that the path can depend on the gene, and are also compatible with more than one path being operational in vivo for a single gene. This work improves our understanding of PIC assembly paths and transcription initiation, and opens perspectives for future modeling and experimental studies.