Frontiers in Systems Biology

AO_SCPLOWBSTRACTC_SCPLOWMathematical modelling and dynamic simulations are commonly used in systems medicine to investigate the interactions between various biological entities in time. The level of model complexity is mainly restricted by the number of model parameters that can be estimated from available experimental data and prior knowledge. The calibration of dynamic models usually requires longitudinal data from multiple individuals, which is challenging to obtain and, consequently, not always available. On the contrary, the collection of cross-sectional data is often more feasible. Here, we demonstrate how the parameters of individual dynamic models can be estimated from such cross-sectional data using a Bayesian updating method. We illustrate this approach on a model for puberty in girls with cross-sectional hormone measurement data.

IntroductionHuman ovaries begin development in utero. Through oogenesis, the numbers of oocytes and primordial follicles peak to a few million during fetal development, then decline to hundreds of thousands per ovary at birth. These primordial follicles do not regenerate and are thus regarded as the ovarian reserve. Over the life course, the reserve continues to deplete, due to atresia and activation, until menopause when about 1000 primordial follicles remain. Exposure to chemotherapy drugs and environmental pollutants can accelerate follicular depletion potentially leading to a greater risk of early menopause, primary ovarian insufficiency (POI), and infertility. Physiologically, the ovarian reserve is depleted in a seemingly biphasic pattern - a slow steady decline from birth to mid-30s, followed by a faster decline to menopause which typically occurs around age 50 years. While this depletion pattern has been described with empirical mathematical formulations, rarely is it modeled mechanistically. A mechanic model that can characterize the dynamics of follicular depletion throughout the life course will help researchers better understand and predict the impact of chemical exposures on ovarian aging. MethodsHere we propose a minimal mechanistic model, which includes (1) a zero-order feedforward inhibition of primordial follicle activation by a local autocrine/paracrine inhibitory factor secreted by the primordial follicles, and (2) a high-gain feedback inhibition of primordial follicle activation by the anti-Mullerian hormone (AMH) secreted by the growing (primary, secondary, and early antral) follicles. The model is configured such that the two regulatory processes prevent primordial follicles from premature overactivation in early and late reproductive life stages, respectively. Two exposure scenarios - chemo-drugs/radiation and tobacco smoke - are presented to demonstrate predictive robustness and biological plausibility of chemically induced increases in cellular atresia. ResultsOur model recapitulates the biphasic depletion curve and predicts a constant supply of growing follicles through most of the active reproductive lifespan. This model predicts that the size of the initial primordial follicle pool plays the most significant role in determining menopausal age and suggests that unilateral ovariectomy may have a more attenuated effect than expected. Simulations of transient exposure to chemotherapy drugs provide an exposure example for promoting atresia of primordial and/or growing follicles and suggest exposure at earlier ages have greater impact on ovarian reserve and menopausal timing than exposure at later ages. Also, simulations of chronic chemical exposures suggest that chemicals which directly promote primordial follicle atresia are more damaging than chemicals directly promoting growing follicle atresia or inhibiting AMH, potentially leading to earlier age at menopause. A specific scenario of chronic exposure to cigarette smoke of various intensities was simulated to validate the prediction power of the model. ConclusionsThe ovary may have compensatory factors to extend reproductive age as long as possible amid insults that reduce the primordial follicle pool. The timing of these insults are likely an important variable. Future elaborations of such mechanistically based computational modeling with integration of in vitro toxicity testing data may help scaling efforts in predicting the implications of reproductive toxicants on ovarian aging.

We recently developed LIONESS, a method to estimate sample-specific networks based on the output of an aggregate network reconstruction approach. In this manuscript, we describe how to apply LIONESS to different network reconstruction algorithms and data types. We highlight how decisions related to data preprocessing may affect the output networks, discuss expected outcomes, and give examples of how to analyze and compare single sample networks.

Computational modeling has emerged as a critical tool in investigating the complex molecular processes involved in biological systems and diseases. In this study, we apply Boolean modeling to uncover the molecular mechanisms underlying Parkinsons disease (PD), one of the most prevalent neurodegenerative disorders. Our approach is based on the PD-map, a comprehensive molecular interaction diagram that captures the key mechanisms involved in the initiation and progression of PD. Using Boolean modeling, we aim to gain a deeper understanding of the disease dynamics, identify potential drug targets, and simulate the response to treatments. Our analysis demonstrates the effectiveness of this approach in uncovering the intricacies of PD. Our results confirm existing knowledge about the disease and provide valuable insights into the underlying mechanisms, ultimately suggesting potential targets for therapeutic intervention. Moreover, our approach allows us to parametrize the models based on omics data for further disease stratification. Our study highlight the value of computational modeling in advancing our understanding of complex biological systems and diseases, emphasizing the importance of continued research in this field. Furthermore, our findings have potential implications for the development of novel therapies for PD, which is a pressing public health concern. Overall, this study represents a significant step forward in the application of computational modeling to the investigation of neurodegenerative diseases, and underscores the power of interdisciplinary approaches in tackling challenging biomedical problems.

1In this study, we investigate the integration of three previously developed tools: FLUTE, VIOLIN, and CLARINET. We show how using these tools together adds additional capabilities in extending models from relevant research literature. We illustrate how we plan to address current modeling pitfalls with these tools (such as machine reading errors and literature volume), and how we plan to use these tools as the foundation for an automated model extension framework. Documentation and links to our tools can be found at: violin-tool.readthedocs.io clarinet-docs.readthedocs.io flute.readthedocs.io

MotivationDue to the utility of knowing the pathway involvement of compounds detected in biological experiments, knowledgebases such as the Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome, and MetaCyc have aggregated pathway annotations of compounds. However, these annotations are largely incomplete and are costly to obtain experimentally and curate from published scientific literature. ResultsWe constructed a new dataset using compounds and their pathway annotations from KEGG, Reactome, and MetaCyc. Using this dataset, we trained and tested an extreme classification model that classifies 8,195 unique pathways based on compound chemical representations with a mean Matthews correlation coefficient (MCC) of 0.9036 {+/-} 0.0033. During model evaluation, we discovered an inconsistency in chemical representations across knowledgebases, which was alleviated by standardizing the chemical representations using InChI (IUPAC International Chemical Identifier) canonicalization. Next, we compared the MCC between compounds and their cross-knowledgebase references. The non-standardized chemical representations had a huge 0.2687 drop in MCC while the standardized chemical representations only had a 0.0384 drop in MCC. Thus, standardizing chemical representation is an essential step when predicting on novel chemical representations. Availability and implementationAll code and data for reproducing the results of this manuscript are available in the following figshare items: Manuscript main results: https://doi.org/10.6084/m9.figshare.28701845 CV analysis of model and dataset of prior studies: https://doi.org/10.6084/m9.figshare.28701590 Contacthunter.moseley@uky.edu Supplementary information<LINK TO SUPPLEMENTAL MATERIAL>

In many fields, research progress may be hindered by indefiniteness of language used to describe experimental conditions and outcomes. Harmonization of data resources generated by independent groups is important for integrative analysis. Adoption of formal ontologies and vocabularies for experiment annotation should help with harmonization tasks, but the use of ontologies also suffers from a lack of definiteness. In this study we explore how natural language characterization of human diseases coupled with ontologic mapping of study outcome terminology can be used to integrate information from multiple studies of genetic origins of disease risk. Open source tools and workflows are presented. This work exposes areas for improvement in tooling for data harmonization, which is a fundamental requirement for efficient research progress.

The network inference problem arises in biological research when one needs to quantitatively choose the best protein-interaction model for explaining a phenotype. The diverse nature of the data and nonlinear dynamics pose significant challenges in the search for the best methodology. In addition to balancing fit and model size, computational efficiency must be considered. Importantly, underlying the measurements, which are affected by experimental noise, there is a complex computational mechanism that is inherently hard to identify. To address these difficulties, we present a novel approach that uses algorithmic complexity to infer a Boolean network model from experimental data. We present an algorithm that is optimal within this framework and allows for asynchronicity network dynamics. Furthermore, we show that using our methodology a solution to the pseudo-time inference problem, which is pertinent to the analysis of single-cell data, can be intertwined with network inference. Results are described for real and simulated datasets.

In this work, our goal was twofold: (1) improve an existing glioblastoma multiforme (GBM) executable mechanistic model and (2) evaluate the effectiveness traditional natural language processing (NLP) pipeline and the generative AI approach in the process of model improvement. We used a suite of graph metrics and tools for interaction filtering and classification to collect data and conduct the analysis. Our results suggest that a more comprehensive literature search is necessary to collect enough information through automated paper retrieval and interaction extraction. Additionally, we found that graph metrics present a promising approach for model refinement, as they can provide useful insights and guidance when selecting new information to be added to a mechanistic model.

In computational modeling, Bounded Linear Temporal Logic (BLTL) is a valuable formalism for describing and verifying the temporal behavior of biological systems. However, translating natural language (NL) descriptions of system behaviors into accurate BLTL properties remains a labor-intensive task, requiring deep expertise in both logic syntax and semantic translation. With the advent of large language models (LLMs), automating this translation has become a promising direction. In this work, we propose an accurate and flexible NL-BLTL transformation framework based on transfer learning. Our approach consists of three stages: 1) Synthetic data generation, where we construct a large-scale NL-BLTL dataset. 2) Pre-training, where we fine-tune LLMs on the synthetic dataset to enhance their ability to characterize logical structure and BLTL specifications. 3) Fine-tuning, where we adapt the pre-trained models to a naive T-cell dataset with manual NL-BLTL annotations. We evaluate the fine-tuned models on the naive T-cell test set and further assess their generalizability on an unseen NL-BLTL dataset in the context of the pancreatic cancer environment, using comprehensive metrics. Experimental results show that models pre-trained on the synthetic data and fine-tuned on real-world annotations outperform both out-of-the-box LLMs, such as GPT-4, and models trained directly on the naive T-cell dataset without pre-training, demonstrating the effectiveness of our framework.

In the study of single cell RNA-seq data, a key component of the analysis is to identify sub-populations of cells in the data. A variety of approaches to this have been considered, and although many machine learning based methods have been developed, these rarely give an estimate of uncertainty in the cluster assignment. To allow for this probabilistic models have been developed, but single cell RNA-seq data exhibit a phenomenon known as dropout, whereby a large proportion of the observed read counts are zero. This poses challenges in developing probabilistic models that appropriately model the data. We develop a novel Dirichlet process mixture model which employs both a mixture at the cell level to model multiple populations of cells, and a zero-inflated negative binomial mixture of counts at the transcript level. By taking a Bayesian approach we are able to model the expression of genes within clusters, and to quantify uncertainty in cluster assignments. It is shown that this approach out-performs previous approaches that applied multinomial distributions to model single cell RNA-seq counts and negative binomial models that do not take into account zero-inflation. Applied to a publicly available data set of single cell RNA-seq counts of multiple cell types from the mouse cortex and hippocampus, we demonstrate how our approach can be used to distinguish sub-populations of cells as clusters in the data, and to identify gene sets that are indicative of membership of a sub-population. The methodology is implemented as an open source Snakemake pipeline available from https://github.com/tt104/scmixture.

In this study, we extend an existing classification method for identifying pairs of genes whose joint expression is associated with binary outcomes to ordinal multi-class outcomes, such as overall survival or disease progression. Our approach is motivated by the need for interpretable classifiers that can provide insights into the underlying biological mechanisms. It can be easily adapted to different research questions, such as identifying gene pair signatures or functional enrichment. We demonstrate that our method is comparable to state-of-the-art classification approaches in terms of performance, while offering the benefit of higher interpretability and adaptability to solve different research questions. Our evaluation on two real-world use cases in glioblastoma and high-grade serous ovarian carcinoma shows that our approach can effectively predict ordinal outcomes and provide novel biological insights. The code is available at https://github.com/oceanefrqt/MBMC.

Published research articles are rich sources of data when the knowledge is incorporated into models. Complex biological systems benefit from computational modelings ability to elucidate dynamics, explain data and address hypotheses. Modeling of pancreatic cancer could guide treatment of this devastating disease that has a known mutational profile disrupting signaling pathways but no reliable therapies. The approach described here is to utilize discrete modeling of the major signaling pathways, metabolism and the tumor microenvironment including macrophages. This modeling approach allows for abstraction in order to assemble large networks to capture numerous facets of the biological system under investigation. The Hallmarks of Cancer are represented as the processes of apoptosis, autophagy, cell cycle progression, inflammation, immune response, oxidative phosphorylation and proliferation. The model is initialized with pancreatic cancer receptors and mutations and simulated in time. The model portrays the hallmarks of cancer and suggests combinations of inhibitors as therapies.

Mathematical modeling of disease and drug action is becoming an indispensable component of drug development, underscored by recent examples of models predicting trial results. To be able to rely on such approaches, decision-makers need to be able to verify those results independently with in silico confirmatory studies. Artificial Intelligence (AI) offers a valuable avenue for improving the reproducibility of complex models, enabling their swift and software-agnostic deployment. This paper highlights AIs impact through a case study on an Ordinary Differential Equation (ODE) model of Retinitis Pigmentosa (RP). We use a version of Chat GPT 4, a sophisticated large language model (LLM) developed by OpenAI, as customized by Mathpix company with additional capabilities. This setup facilitated the extraction of equations from the PDF and converted into a human-readable, text-based definition language called Antimony, which is part of the Python package tellurium. Subsequently, the model was converted into Systems Biology Markup Language (SBML) using tellurium and uploaded onto the jink[o] platform for simulation. The RP model was efficiently and accurately implemented using AI techniques. Furthermore, we were able to reproduce the model behavior presented in the literature. Our findings advocate for the broader application of AI in mathematical model re-implementations to ensure reliability and reproducibility of the results.

Neural Mass Models (NMMs) are important mathematical tools for inferring hidden neural mechanisms that generate healthy and pathological brain activities. A critical step in the inference process is parameter estimation, which calibrates NMMs based on measured neuroimaging data. While parameter estimation can be conducted via various approaches, one of the most influential methods is Dynamic Causal Modelling (DCM). DCM adopts a Bayesian inference approach that relies on, and is sensitive to, the specification of prior parameter distributions reflecting a priori hypotheses about the causes of data. However, most parameters of NMMs encode neuronal properties that are not directly measurable. For this reason, in the absence of sufficient empirical data and well-founded prior beliefs, inference becomes increasingly susceptible to bias. Therefore, it was imperative to establish a comprehensive strategy for mapping model parameters to data. This study proposes a computational extension of DCM, named DCM with dynamics-informed priors (DIP-DCM), which adopts a genetic algorithm (GA) to map parameter values to model dynamics. Optimal sub-regions of the parameter space were subsequently selected and translated into groups of parameter priors for DCM. DIP-DCM was compared to the standard DCM inference and to the standalone GA, using two independent neuroimaging datasets. Results indicated that DIP-DCM models were the best predictors of data and captured key mechanistic signatures of psychiatric disease and pharmacological interventions. Overall, DIP-DCM addressed degeneracy, handled local minima, and explored diverse parameter regimes following trajectories informed directly by model dynamics and data. This study suggests that DIP-DCM is an advantageous route to parameter estimation when information is limited, enabling a data-driven derivation of parameter priors - or hypotheses - in exploratory studies, across different biological contexts and datasets.

Gene functional enrichment is a mainstay of genomics, but it relies on manually curated databases of gene functions that are incomplete and unaware of the biological context. Here we present an alternative machine learning approach, Deep Functional Synthesis (DeepSyn), which moves beyond gene function databases to dynamically infer the functions of a gene set from its associated network of literature and data, conditioned on the disease and drug context of the current experiment. Using a knowledge graph with 3,048,803 associations between genes, diseases, drugs, and functions, DeepSyn obtained accurate performance (range 0.74 AUC to 0.96 AUC) on a variety of biological applications including drug target identification, gene set functional enrichment, and disease gene prediction. AvailabilityThe DeepSyn codebase is available on GitHub at http://github.com/wangshenguiuc/DeepSyn/ under an open source distribution license.

MotivationIdentifying cancer genes remains a significant challenge in cancer genomics research. Annotated gene sets encode functional associations among multiple genes, and cancer genes have been shown to cluster in hallmark signaling pathways and biological processes. The knowledge of annotated gene sets is critical for discovering cancer genes but remains to be fully exploited. ResultsHere, we present the DIsease-Specific Hypergraph neural network (DISHyper), a hypergraph-based computational method that integrates the knowledge from multiple types of annotated gene sets to predict cancer genes. First, our benchmark results demonstrate that DISHyper outperforms the existing state-of-the-art methods and highlight the advantages of employing hypergraphs for representing annotated gene sets. Second, we validate the accuracy of DISHyper-predicted cancer genes using functional validation results and multiple independent functional genomics data. Third, our model predicts 44 novel cancer genes, and subsequent analysis shows their significant associations with multiple types of cancers. Overall, our study provides a new perspective for discovering cancer genes and reveals previously undiscovered cancer genes. AvailabilityDISHyper is freely available for download at https://github.com/genemine/DISHyper. Contactjxwang@mail.csu.edu.cn

The pathway curation task involves analyzing scientific literature to identify and represent cellular processes as pathways. This process, often time-consuming and labor-intensive, requires significant curation efforts amidst the rapidly growing biomedical literature. Natural Language Processing (NLP) offers a promising method to automatically extract these interactions from scientific texts. Despite immense progress, there remains room for improvement in these systems. The emergence of Large Language Models (LLMs) provides a promising solution for this challenge. Our study conducts a preliminary investigation into leveraging LLMs for the pathway curation task. This paper first presents a review of the current state-of-the-art algorithms for the pathway curation task. Our objective is to check the feasibility and formulate strategies of using these LLMs to improve the accuracy of pathway curation task. Our experiments demonstrate that our GPT-3.5 based fine-tuned models outperforms existing state-of-the-art methods. Specifically, our model achieved a 10 basis point improvement in over-all recall and F1 score compared to the best existing algorithms. These findings highlight the potential of LLMs in pathway curation tasks, warranting further research and substantial efforts in this direction. Keypoints/ObjectivesO_LIStudy evaluates the feasibility of using Large Language Models (LLMs) for pathway curation in scientific literature. C_LIO_LIUsing GPT-3.5 based fine tuned models for pathway curation, we compare its performance with existing methods, focusing on precision, recall and F1 score metrics. C_LIO_LIEmphasize the promise and need for further research on using LLMs for pathway curation. C_LI

Annotating genes is essential to crop development, and understanding gene functions sheds light on developing crop improvement strategies, such as marker-assisted breeding, genetic modification, or pest resistance. Through an extensive experimental effort and computational annotation projection, tens of thousands of genes have been annotated across plant species, with most of the gene annotations focusing on a well-studied species, Arabidopsis thaliana, but this represents a small fraction of the hundreds of thousands of genes across these different plant species. Phenotypes and their traits result from multiple processes and events involving multiscale information encoded from different omics, such as genomes, proteomes, or transcriptomes. This stresses a need for an efficient computational approach to capture and integrate information from biological networks and transfer this knowledge from well-studied species to unknown species to annotate and discover functional relationships between annotations and genes. Despite recent progress, existing methods only consider one or a few omics levels to perform reasoning on functional annotation-to-gene relations. The main objective of this study is to generate and explore a large-scale plant biological knowledge graph, the DasDB, and to enrich gene functional annotation linked to genes in different species using graph neural networks (GNNs). Integrating various data sources from different omics has resulted in a comprehensive graph database, facilitating researchers in-depth understanding of complex biological networks at the highest level. In addition, applying GNNs on a large-scale knowledge graph database has shown promise in the ability of deep learning models to transfer this information from well-studied plant species to less-characterized plant species, outperforming the transfer of information done using only orthology relationships. This study benchmarks a new research direction in producing new functional annotation discovery in plant species with limited functional annotations. This pipeline was applied to a specific research problem: the mechanism involved in pea nodule nitrogen fixation. We identified known gene markers of this process through a systematic analysis of the DasDB, showing the relevance of our approach. Furthermore, new potential targets to better understand and improve this process were identified.

Background The effectiveness of many targeted therapies is limited by toxicity and the rise of drug resistance. A growing appreciation of the inherent redundancies of cancer signaling has led to a rise in the number of combination therapies under development, but a better understanding of the overall cancer network topology would provide a conceptual framework for choosing effective combination partners. In this work, we explore the scale-free nature of cancer protein-protein interaction networks in 14 indications. Scale-free networks, characterized by a power-law degree distribution, are known to be resilient to random attack on their nodes, yet vulnerable to directed attacks on their hubs (their most highly connected nodes).Results Consistent with the properties of scale-free networks, we find that lethal genes are associated with ∼5-fold higher protein connectivity partners than non-lethal genes. This provides a biological rationale for a hub-centered combination attack. Our simulations show that combinations targeting hubs can efficiently disrupt 50% of network integrity by inhibiting less than 1% of the connected proteins, whereas a random attack can require inhibition of more than 30% of the connected proteins.Conclusions We find that the scale-free nature of cancer networks makes them vulnerable to focused attack on their highly connected protein hubs. Thus, we propose a new strategy for designing combination therapies by targeting hubs in cancer networks that are not associated with relevant toxicity networks.Competing Interest StatementThe authors have declared no competing interest.View Full Text