Experimental validation of computerised models of clustering of platelet glycoprotein receptors that signal via tandem SH2 domain proteins.

The clustering of platelet glycoprotein receptors with cytosolic YxxL and YxxM motifs, including GPVI, CLEC-2 and PEAR1, triggers activation via phosphorylation of the conserved tyrosine residues and recruitment of the tandem SH2 (Src homology 2) domain effector proteins, Syk and PI 3-kinase. We have modelled the clustering of these receptors with monovalent, divalent and tetravalent soluble ligands and with transmembrane ligands based on the law of mass action using ordinary differential equations and agent-based modelling. The models were experimentally evaluated in platelets and transfected cell lines using monovalent and multivalent ligands, including novel nanobody-based divalent and tetravalent ligands, by fluorescence correlation spectroscopy. Ligand valency, receptor number, receptor dimerisation, receptor phosphorylation and a cytosolic tandem SH2 domain protein act in synergy to drive receptor clustering. Threshold concentrations of a CLEC-2-blocking antibody and Syk inhibitor act in synergy to block platelet aggregation. This offers a strategy for countering the effect of avidity of multivalent ligands and in limiting off-target effects.

A framework for validating AI in precision medicine: considerations from the European ITFoC consortium.

BACKGROUND: Artificial intelligence (AI) has the potential to transform our healthcare systems significantly. New AI technologies based on machine learning approaches should play a key role in clinical decision-making in the future. However, their implementation in health care settings remains limited, mostly due to a lack of robust validation procedures. There is a need to develop reliable assessment frameworks for the clinical validation of AI. We present here an approach for assessing AI for predicting treatment response in triple-negative breast cancer (TNBC), using real-world data and molecular -omics data from clinical data warehouses and biobanks. METHODS: The European "ITFoC (Information Technology for the Future Of Cancer)" consortium designed a framework for the clinical validation of AI technologies for predicting treatment response in oncology. RESULTS: This framework is based on seven key steps specifying: (1) the intended use of AI, (2) the target population, (3) the timing of AI evaluation, (4) the datasets used for evaluation, (5) the procedures used for ensuring data safety (including data quality, privacy and security), (6) the metrics used for measuring performance, and (7) the procedures used to ensure that the AI is explainable. This framework forms the basis of a validation platform that we are building for the "ITFoC Challenge". This community-wide competition will make it possible to assess and compare AI algorithms for predicting the response to TNBC treatments with external real-world datasets. CONCLUSIONS: The predictive performance and safety of AI technologies must be assessed in a robust, unbiased and transparent manner before their implementation in healthcare settings. We believe that the consideration of the ITFoC consortium will contribute to the safe transfer and implementation of AI in clinical settings, in the context of precision oncology and personalized care.

Synergistic cross-talk of hedgehog and interleukin-6 signaling drives growth of basal cell carcinoma.

Persistent activation of hedgehog (HH)/GLI signaling accounts for the development of basal cell carcinoma (BCC), a very frequent nonmelanoma skin cancer with rising incidence. Targeting HH/GLI signaling by approved pathway inhibitors can provide significant therapeutic benefit to BCC patients. However, limited response rates, development of drug resistance, and severe side effects of HH pathway inhibitors call for improved treatment strategies such as rational combination therapies simultaneously inhibiting HH/GLI and cooperative signals promoting the oncogenic activity of HH/GLI. In this study, we identified the interleukin-6 (IL6) pathway as a novel synergistic signal promoting oncogenic HH/GLI via STAT3 activation. Mechanistically, we provide evidence that signal integration of IL6 and HH/GLI occurs at the level of cis-regulatory sequences by co-binding of GLI and STAT3 to common HH-IL6 target gene promoters. Genetic inactivation of Il6 signaling in a mouse model of BCC significantly reduced in vivo tumor growth by interfering with HH/GLI-driven BCC proliferation. Our genetic and pharmacologic data suggest that combinatorial HH-IL6 pathway blockade is a promising approach to efficiently arrest cancer growth in BCC patients.

Network and systems biology: essential steps in virtualising drug discovery and development.

The biological processes that keep us healthy or cause disease, as well as the mechanisms of action of possible drugs are inherently complex. In the face of this complexity, attempts at discovering new drugs to treat diseases have alternated between trial-and-error (typically on experimental systems) and grand simplification, usually based on much too little information. We now have the chance to combine these strategies through establishment of 'virtual patient' models, centred on a detailed molecular characterisation of thousands or even, in the future, millions of patients. In doing so, we lay the foundations for truly personalised therapy, as well as a far-reaching virtualisation of drug discovery and development in oncology and other areas of medicine.

ConsensusPathDB: toward a more complete picture of cell biology.

ConsensusPathDB is a meta-database that integrates different types of functional interactions from heterogeneous interaction data resources. Physical protein interactions, metabolic and signaling reactions and gene regulatory interactions are integrated in a seamless functional association network that simultaneously describes multiple functional aspects of genes, proteins, complexes, metabolites, etc. With 155,432 human, 194,480 yeast and 13,648 mouse complex functional interactions (originating from 18 databases on human and eight databases on yeast and mouse interactions each), ConsensusPathDB currently constitutes the most comprehensive publicly available interaction repository for these species. The Web interface at http://cpdb.molgen.mpg.de offers different ways of utilizing these integrated interaction data, in particular with tools for visualization, analysis and interpretation of high-throughput expression data in the light of functional interactions and biological pathways.

An agent-based approach for modelling and simulation of glycoprotein VI receptor diffusion, localisation and dimerisation in platelet lipid rafts.

Receptor diffusion plays an essential role in cellular signalling via the plasma membrane microenvironment and receptor interactions, but the regulation is not well understood. To aid in understanding of the key determinants of receptor diffusion and signalling, we developed agent-based models (ABMs) to explore the extent of dimerisation of the platelet- and megakaryocyte-specific receptor for collagen glycoprotein VI (GPVI). This approach assessed the importance of glycolipid enriched raft-like domains within the plasma membrane that lower receptor diffusivity. Our model simulations demonstrated that GPVI dimers preferentially concentrate in confined domains and, if diffusivity within domains is decreased relative to outside of domains, dimerisation rates are increased. While an increased amount of confined domains resulted in further dimerisation, merging of domains, which may occur upon membrane rearrangements, was without effect. Modelling of the proportion of the cell membrane which constitutes lipid rafts indicated that dimerisation levels could not be explained by these alone. Crowding of receptors by other membrane proteins was also an important determinant of GPVI dimerisation. Together, these results demonstrate the value of ABM approaches in exploring the interactions on a cell surface, guiding the experimentation for new therapeutic avenues.

Prediction in the face of uncertainty: a Monte Carlo-based approach for systems biology of cancer treatment.

Cancer is known to be a complex disease and its therapy is difficult. Much information is available on molecules and pathways involved in cancer onset and progression and this data provides a valuable resource for the development of predictive computer models that can help to identify new potential drug targets or to improve therapies. Modeling cancer treatment has to take into account many cellular pathways usually leading to the construction of large mathematical models. The development of such models is complicated by the fact that relevant parameters are either completely unknown, or can at best be measured under highly artificial conditions. Here we propose an approach for constructing predictive models of such complex biological networks in the absence of accurate knowledge on parameter values, and apply this strategy to predict the effects of perturbations induced by anti-cancer drug target inhibitions on an epidermal growth factor (EGF) signaling network. The strategy is based on a Monte Carlo approach, in which the kinetic parameters are repeatedly sampled from specific probability distributions and used for multiple parallel simulations. Simulation results from different forms of the model (e.g., a model that expresses a certain mutation or mutation pattern or the treatment by a certain drug or drug combination) can be compared with the unperturbed control model and used for the prediction of the perturbation effects. This framework opens the way to experiment with complex biological networks in the computer, likely to save costs in drug development and to improve patient therapy.

Mini-batch optimization enables training of ODE models on large-scale datasets.

Quantitative dynamic models are widely used to study cellular signal processing. A critical step in modelling is the estimation of unknown model parameters from experimental data. As model sizes and datasets are steadily growing, established parameter optimization approaches for mechanistic models become computationally extremely challenging. Mini-batch optimization methods, as employed in deep learning, have better scaling properties. In this work, we adapt, apply, and benchmark mini-batch optimization for ordinary differential equation (ODE) models, thereby establishing a direct link between dynamic modelling and machine learning. On our main application example, a large-scale model of cancer signaling, we benchmark mini-batch optimization against established methods, achieving better optimization results and reducing computation by more than an order of magnitude. We expect that our work will serve as a first step towards mini-batch optimization tailored to ODE models and enable modelling of even larger and more complex systems than what is currently possible.

The structural impact of cancer-associated missense mutations in oncogenes and tumor suppressors.

BACKGROUND: Current large-scale cancer sequencing projects have identified large numbers of somatic mutations covering an increasing number of different cancer tissues and patients. However, the characterization of these mutations at the structural and functional level remains a challenge. RESULTS: We present results from an analysis of the structural impact of frequent missense cancer mutations using an automated method. We find that inactivation of tumor suppressors in cancer correlates frequently with destabilizing mutations preferably in the core of the protein, while enhanced activity of oncogenes is often linked to specific mutations at functional sites. Furthermore, our results show that this alteration of oncogenic activity is often associated with mutations at ATP or GTP binding sites. CONCLUSIONS: With our findings we can confirm and statistically validate the hypotheses for the gain-of-function and loss-of-function mechanisms of oncogenes and tumor suppressors, respectively. We show that the distinct mutational patterns can potentially be used to pre-classify newly identified cancer-associated genes with yet unknown function.

ConsensusPathDB--a database for integrating human functional interaction networks.

ConsensusPathDB is a database system for the integration of human functional interactions. Current knowledge of these interactions is dispersed in more than 200 databases, each having a specific focus and data format. ConsensusPathDB currently integrates the content of 12 different interaction databases with heterogeneous foci comprising a total of 26,133 distinct physical entities and 74,289 distinct functional interactions (protein-protein interactions, biochemical reactions, gene regulatory interactions), and covering 1738 pathways. We describe the database schema and the methods used for data integration. Furthermore, we describe the functionality of the ConsensusPathDB web interface, where users can search and visualize interaction networks, upload, modify and expand networks in BioPAX, SBML or PSI-MI format, or carry out over-representation analysis with uploaded identifier lists with respect to substructures derived from the integrated interaction network. The ConsensusPathDB database is available at: http://cpdb.molgen.mpg.de.

GeNGe: systematic generation of gene regulatory networks.

UNLABELLED: The analysis of gene regulatory networks (GRNs) is a central goal of bioinformatics highly accelerated by the advent of new experimental techniques, such as RNA interference. A battery of reverse engineering methods has been developed in recent years to reconstruct the underlying GRNs from these and other experimental data. However, the performance of the individual methods is poorly understood and validation of algorithmic performances is still missing to a large extent. To enable such systematic validation, we have developed the web application GeNGe (GEne Network GEnerator), a controlled framework for the automatic generation of GRNs. The theoretical model for a GRN is a non-linear differential equation system. Networks can be user-defined or constructed in a modular way with the option to introduce global and local network perturbations. Resulting data can be used, e.g. as benchmark data for evaluating GRN reconstruction methods or for predicting effects of perturbations as theoretical counterparts of biological experiments. AVAILABILITY: Available online at http://genge.molgen.mpg.de. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

PROTEOMER: A workflow-optimized laboratory information management system for 2-D electrophoresis-centered proteomics.

In recent years proteomics became increasingly important to functional genomics. Although a large amount of data is generated by high throughput large-scale techniques, a connection of these mostly heterogeneous data from different analytical platforms and of different experiments is limited. Data mining procedures and algorithms are often insufficient to extract meaningful results from large datasets and therefore limit the exploitation of the generated biological information. In our proteomic core facility, which almost exclusively focuses on 2-DE/MS-based proteomics, we developed a proteomic database custom tailored to our needs aiming at connecting MS protein identification information to 2-DE derived protein expression profiles. The tools developed should not only enable an automatic evaluation of single experiments, but also link multiple 2-DE experiments with MS-data on different levels and thereby helping to create a comprehensive network of our proteomics data. Therefore the key feature of our "PROTEOMER" database is its high cross-referencing capacity, enabling integration of a wide range of experimental data. To illustrate the workflow and utility of the system, two practical examples are provided to demonstrate that proper data cross-referencing can transform information into biological knowledge.

Efficient Parameter Estimation Enables the Prediction of Drug Response Using a Mechanistic Pan-Cancer Pathway Model.

Mechanistic models are essential to deepen the understanding of complex diseases at the molecular level. Nowadays, high-throughput molecular and phenotypic characterizations are possible, but the integration of such data with prior knowledge on signaling pathways is limited by the availability of scalable computational methods. Here, we present a computational framework for the parameterization of large-scale mechanistic models and its application to the prediction of drug response of cancer cell lines from exome and transcriptome sequencing data. This framework is over 104 times faster than state-of-the-art methods, which enables modeling at previously infeasible scales. By applying the framework to a model describing major cancer-associated pathways (>1,200 species and >2,600 reactions), we could predict the effect of drug combinations from single drug data. This is the first integration of high-throughput datasets using large-scale mechanistic models. We anticipate this to be the starting point for development of more comprehensive models allowing a deeper mechanistic insight.

Models of Models: A Translational Route for Cancer Treatment and Drug Development.

Every patient and every disease is different. Each patient therefore requires a personalized treatment approach. For technical reasons, a personalized approach is feasible for treatment strategies such as surgery, but not for drug-based therapy or drug development. The development of individual mechanistic models of the disease process in every patient offers the possibility of attaining truly personalized drug-based therapy and prevention. The concept of virtual clinical trials and the integrated use of in silico, in vitro, and in vivo models in preclinical development could lead to significant gains in efficiency and order of magnitude increases in the cost effectiveness of drug development and approval. We have developed mechanistic computational models of large-scale cellular signal transduction networks for prediction of drug effects and functional responses, based on patient-specific multi-level omics profiles. However, a major barrier to the use of such models in a clinical and developmental context is the reliability of predictions. Here we detail how the approach of using "models of models" has the potential to impact cancer treatment and drug development. We describe the iterative refinement process that leverages the flexibility of experimental systems to generate highly dimensional data, which can be used to train and validate computational model parameters and improve model predictions. In this way, highly optimized computational models with robust predictive capacity can be generated. Such models open up a number of opportunities for cancer drug treatment and development, from enhancing the design of experimental studies, reducing costs, and improving animal welfare, to increasing the translational value of results generated.

Molecular dissection of colorectal cancer in pre-clinical models identifies biomarkers predicting sensitivity to EGFR inhibitors.

Colorectal carcinoma represents a heterogeneous entity, with only a fraction of the tumours responding to available therapies, requiring a better molecular understanding of the disease in precision oncology. To address this challenge, the OncoTrack consortium recruited 106 CRC patients (stages I-IV) and developed a pre-clinical platform generating a compendium of drug sensitivity data totalling >4,000 assays testing 16 clinical drugs on patient-derived in vivo and in vitro models. This large biobank of 106 tumours, 35 organoids and 59 xenografts, with extensive omics data comparing donor tumours and derived models provides a resource for advancing our understanding of CRC. Models recapitulate many of the genetic and transcriptomic features of the donors, but defined less complex molecular sub-groups because of the loss of human stroma. Linking molecular profiles with drug sensitivity patterns identifies novel biomarkers, including a signature outperforming RAS/RAF mutations in predicting sensitivity to the EGFR inhibitor cetuximab.

Predictive Modeling of Drug Treatment in the Area of Personalized Medicine.

Despite a growing body of knowledge on the mechanisms underlying the onset and progression of cancer, treatment success rates in oncology are at best modest. Current approaches use statistical methods that fail to embrace the inherent and expansive complexity of the tumor/patient/drug interaction. Computational modeling, in particular mechanistic modeling, has the power to resolve this complexity. Using fundamental knowledge on the interactions occurring between the components of a complex biological system, large-scale in silico models with predictive capabilities can be generated. Here, we describe how mechanistic virtual patient models, based on systematic molecular characterization of patients and their diseases, have the potential to shift the theranostic paradigm for oncology, both in the fields of personalized medicine and targeted drug development. In particular, we highlight the mechanistic modeling platform ModCell™ for individualized prediction of patient responses to treatment, emphasizing modeling techniques and avenues of application.

Multi-omic profiles of human non-alcoholic fatty liver disease tissue highlight heterogenic phenotypes.

Non-alcoholic fatty liver disease (NAFLD) is a consequence of sedentary life style and high fat diets with an estimated prevalence of about 30% in western countries. It is associated with insulin resistance, obesity, glucose intolerance and drug toxicity. Additionally, polymorphisms within, e.g., APOC3, PNPLA3, NCAN, TM6SF2 and PPP1R3B, correlate with NAFLD. Several studies have already investigated later stages of the disease. This study explores the early steatosis stage of NAFLD with the aim of identifying molecular mechanisms underlying the etiology of NAFLD. We analyzed liver biopsies and serum samples from patients with high- and low-grade steatosis (also pre-disease states) employing transcriptomics, ELISA-based serum protein analyses and metabolomics. Here, we provide a detailed description of the various related datasets produced in the course of this study. These datasets may help other researchers find new clues for the etiology of NAFLD and the mechanisms underlying its progression to more severe disease states.

Simulation of DNA array hybridization experiments and evaluation of critical parameters during subsequent image and data analysis.

BACKGROUND: Gene expression analyses based on complex hybridization measurements have increased rapidly in recent years and have given rise to a huge amount of bioinformatic tools such as image analyses and cluster analyses. However, the amount of work done to integrate and evaluate these tools and the corresponding experimental procedures is not high. Although complex hybridization experiments are based on a data production pipeline that incorporates a significant amount of error parameters, the evaluation of these parameters has not been studied yet in sufficient detail. RESULTS: In this paper we present simulation studies on several error parameters arising in complex hybridization experiments. A general tool was developed that allows the design of exactly defined hybridization data incorporating, for example, variations of spot shapes, spot positions and local and global background noise. The simulation environment was used to judge the influence of these parameters on subsequent data analysis, for example image analysis and the detection of differentially expressed genes. As a guide for simulating expression data real experimental data were used and model parameters were adapted to these data. Our results show how measurement error can be balanced by the analysis tools. CONCLUSIONS: We describe an implemented model for the simulation of DNA-array experiments. This tool was used to judge the influence of critical parameters on the subsequent image analysis and differential expression analysis. Furthermore the tool can be used to guide future experiments and to improve performance by better experimental design. Series of simulated images varying specific parameters can be downloaded from our web-site: http://www.molgen.mpg.de/~lh_bioinf/projects/simulation/biotech/

Inferring dynamic properties of biochemical reaction networks from structural knowledge.

Functional properties of biochemical networks depend on both the network structure and the kinetic parameters. Extensive data on metabolic network topologies have been collected in databases, but much less information is available about the kinetic constants or metabolite concentrations. Depending on the values of these parameters, metabolic fluxes and control coefficients may vary within a wide range. Nevertheless, some of the parameters may have little influence on the observables of interest. We address the question whether, despite uncertainty about kinetic parameters, probabilistic statements can be made about dynamic network features. To this end, we perform a variability analysis of the parameters: assuming that the parameters follow statistical distributions, we compute the resulting distributions of the network properties like metabolic fluxes, concentrations, or control coefficients by Monte Carlo simulation. In this manner, we study systematically the possible distributions arising from typical topologies of biochemical networks such as linear chains, branched networks, and signaling and gene expression cascades. This analysis reveals how much information about dynamic behavior can be drawn from structural knowledge.