Carolina Alvarez-Garavito

Mechanistic models provide a framework for describing, understanding, and predicting biological processes, but their development requires an interdisciplinary team with expertise in biology, mathematics, and computation. Here we present Octomod, a multi-agent workflow based on large language models (LLMs) that integrates biological knowledge retrieval, model construction, parameter estimation, and hypothesis testing. Built on established software tools, including AMICI and pyPESTO, and community standards, including SBML and PEtab, Octomod supports model development and refinement through natural-language interaction with a user. The agents' proposals must pass defined checks and the user's decision before they are used. Every input, model, check, and decision is recorded, so that model building remains transparent and reproducible. We applied Octomod to four problems. Three used published data and differ in model size and in how much the user guided the run: Octomod constructed a model that matched a published model without user guidance, supported the interactive refinement of a gene-regulatory model, and built models of cell-cell communication that explained the responses of five immune cell types to different stimuli. In the fourth application, on newly generated and unpublished lipidomics data, Octomod helped to determine which processes the measurements could inform, identify unresolved label fluxes, and reduce a supplied model to those processes. Together, these results show how LLMs can strengthen and complement established workflows for constructing and refining biologically informed models.

The longitudinal immunological mechanisms driving the heterogeneous clinical phenotypes of Long COVID (LC) remain poorly understood. Here, we quantified 54 circulating immune biomarkers in 355 individuals followed for up to 24 months post-infection. Using joint factor analysis and Bayesian structural equation modelling, we demonstrate that despite clinical heterogeneity, the respiratory, chronic pain, and chronic fatigue-like LC phenotypes share a convergent immune signature. This signature is characterized by persistent upregulation of the Th17-associated cytokine IL-17F from 3 to 12 months. Early recovery phases were marked by down-regulated angiogenic factors, which later transitioned to sustained elevation of VEGF-C, fractalkine, eotaxin, and CTACK, indicating dysregulated vascular repair. While the neurosensorial phenotype lacked systemic immune perturbations, the chronic fatigue-like phenotype demonstrated a distinct, prolonged trajectory of active immune dysregulation driven by self-sustaining tissue remodelling rather than generalized hyperinflammation. These findings provide a longitudinal framework for understanding LC pathogenesis, identifying maladaptive immune-remodelling pathways as potential therapeutic targets in post-acute infection syndromes.

Post-Acute Infectious Syndromes (PAIS) refer to the symptoms persisting months after initial infection. Clinical research studies on this topic often collect rich, multi-modal datasets. Yet, the complexity of the datasets and the lack of a precise clinical case definition pose difficulties in creating comprehensive analyses. Here, we present a generalisable framework for analysing data from longitudinal studies of PAIS using Latent Transition Analysis (LTA). It enables the identification of disease phenotypes and the patient-level analysis of transitions between them, without relying on predefined clinical categorisations. Furthermore, we introduce a method for incorporating covariate information, which enables exploration of how patient characteristics influence disease trajectories. We apply this methodology to the ORCHESTRA dataset, composed of individuals affected by SARS-CoV-2 infection from multiple European centres, for investigation into Post-COVID-19 condition (PCC). 5094 patient assessments were collected at SARS-CoV-2 infection, and at 6, 12, 18, and 24 months of follow-up. Our model identifies distinct PCC phenotypes with patient trajectories impacted by age and sex. Our results highlight how LTA can enhance the interpretability of complex, time-resolved clinical data, support personalized patient monitoring and management, and accelerate therapeutic development for other PAISs, too.

Background: Persistent low-grade inflammation has been proposed as part of the biological mechanisms underlying post-COVID condition (PCC), which can result in laboratory tests abnormalities. However, the accuracy of routine laboratory tests for the diagnosis and follow-up of PCC is still under discussion. Methods: Patients with SARS-CoV-2 infection enrolled in the prospective, multinational ORCHESTRA cohort study, which included both European and non-European countries, were followed up for 18 months after acute infection. Blood test results were collected at acute infection and at 6, 12, and 18 months. A multivariable analysis was performed to estimate the relationship between the alterations of biochemical markers and the presence of four distinct PCC phenotypes, identified previously through a principal component analysis—respiratory (RESc), chronic pain (CPc), chronic fatigue (CFc), and neurosensorial (NSc)—during follow-up. Furthermore, this study investigated the correlation between biochemical parameters measured during the acute phase and the subsequent development of PCC. Finally, the relationship between the severity of the acute infection and biochemical abnormalities observed during follow-up was assessed. Results: The cohort included 4587 patients, 58% male, with a mean age of 58.7 (±15.5) years. A robust multivariable analysis demonstrated that, compared to controls, patients with PCC, and in particular those in the RESc cluster, presented higher mean C-reactive protein (CRP) levels at the 12- and 18-month follow-up (p-value = 0.01). In each follow-up, CRP values in patients with PCC and RESc were above 3 mg/L, corresponding to those observed in low-grade inflammation (3–10 mg/L). The severity of COVID-19 acute infection was associated with increased levels of CRP, ferritin and LDH during follow-up (p < 0.001). Biochemistry abnormalities detected during the early stages of acute COVID-19 did not correlate with an increased risk of developing PCC and its phenotypes. Conclusions: In patients with the RESc PCC phenotype, identified through a principal component analysis, blood test abnormalities consistent with prolonged and sustained low-grade inflammation can be detected up to 18 months after acute infection, supporting its role in the pathogenesis of PCC. Based on these results, trials on anti-inflammatory drugs, together with symptom-tailored interventions for patients with RESc, should be planned to prove their effectiveness in managing PCC and improving patient outcomes.

Metabolic responses to dietary interventions often show high inter-individual variations, but the factors explaining these variations as well as underlying mechanisms and their interrelation remain largely unknown. Knowing which factors are relevant for the individual response to a dietary intervention and to which extent, is essential to provide targeted dietary recommendations. Therefore, in persons at increased cardiometabolic risk, a 6 week RCT comparing the individual effects of three dietary patterns was conducted. We observed highly individualized effects in the cholesterol-lowering ability of the dietary patterns, which seem mainly related to differences in the lipid and metabolic profile of the participants in addition to diet-specific characteristics, while the gut microbiome and the polygenetic risk for hyperlipidemia modulate the effects. Moreover, the individual response to the interventions might be predicted by machine-learning models providing multi-omics signatures that could be applied as biomarkers for stratification of persons to specific dietary interventions for the optimization of personalized dietary approaches.