Bridging scales in cancer progression: mapping genotype to phenotype using neural networks

Modelling the Tumour Microenvironment, but What Exactly Do We Mean by "Model"?

The Oxford English Dictionary includes 17 definitions for the word "model" as a noun and another 11 as a verb. Therefore, context is necessary to understand the meaning of the word model. For instance, "model railways" refer to replicas of railways and trains at a smaller scale and a "model student" refers to an exemplary individual. In some cases, a specific context, like cancer research, may not be sufficient to provide one specific meaning for model. Even if the context is narrowed, specifically, to research related to the tumour microenvironment, "model" can be understood in a wide variety of ways, from an animal model to a mathematical expression. This paper presents a review of different "models" of the tumour microenvironment, as grouped by different definitions of the word into four categories: model organisms, in vitro models, mathematical models and computational models. Then, the frequencies of different meanings of the word "model" related to the tumour microenvironment are measured from numbers of entries in the MEDLINE database of the United States National Library of Medicine at the National Institutes of Health. The frequencies of the main components of the microenvironment and the organ-related cancers modelled are also assessed quantitatively with specific keywords. Whilst animal models, particularly xenografts and mouse models, are the most commonly used "models", the number of these entries has been slowly decreasing. Mathematical models, as well as prognostic and risk models, follow in frequency, and these have been growing in use.

Multiscale modeling meets machine learning: What can we learn?

Machine learning is increasingly recognized as a promising technology in the biological, biomedical, and behavioral sciences. There can be no argument that this technique is incredibly successful in image recognition with immediate applications in diagnostics including electrophysiology, radiology, or pathology, where we have access to massive amounts of annotated data. However, machine learning often performs poorly in prognosis, especially when dealing with sparse data. This is a field where classical physics-based simulation seems to remain irreplaceable. In this review, we identify areas in the biomedical sciences where machine learning and multiscale modeling can mutually benefit from one another: Machine learning can integrate physics-based knowledge in the form of governing equations, boundary conditions, or constraints to manage ill-posted problems and robustly handle sparse and noisy data; multiscale modeling can integrate machine learning to create surrogate models, identify system dynamics and parameters, analyze sensitivities, and quantify uncertainty to bridge the scales and understand the emergence of function. With a view towards applications in the life sciences, we discuss the state of the art of combining machine learning and multiscale modeling, identify applications and opportunities, raise open questions, and address potential challenges and limitations. We anticipate that it will stimulate discussion within the community of computational mechanics and reach out to other disciplines including mathematics, statistics, computer science, artificial intelligence, biomedicine, systems biology, and precision medicine to join forces towards creating robust and efficient models for biological systems.

Histoecology: Applying Ecological Principles and Approaches to Describe and Predict Tumor Ecosystem Dynamics Across Space and Time

Cancer cells exist within a complex spatially structured ecosystem composed of resources and different cell types. As the selective pressures imposed by this environment determine the fate of cancer cells, an improved understanding of how this ecosystem evolves will better elucidate how tumors grow and respond to therapy. State of the art imaging methods can now provide highly resolved descriptions of the microenvironment, yielding the data required for a thorough study of its role in tumor growth and treatment resistance. The field of landscape ecology has been studying such species-environment relationship for decades, and offers many tools and perspectives that cancer researchers could greatly benefit from. Here, we discuss one such tool, species distribution modeling (SDM), that has the potential to, among other things, identify critical environmental factors that drive tumor evolution and predict response to therapy. SDMs only scratch the surface of how ecological theory and methods can be applied to cancer, and we believe further integration will take cancer research in exciting new and productive directions. Significance: Here we describe how species distribution modeling can be used to quantitatively describe the complex relationship between tumor cells and their microenvironment. Such a description facilitates a deeper understanding of cancers eco-evolutionary dynamics, which in turn sheds light on the factors that drive tumor growth and response to treatment.

Stability of spontaneous, correlated activity in mouse auditory cortex

Neural systems can be modeled as complex networks in which neural elements are represented as nodes linked to one another through structural or functional connections. The resulting network can be analyzed using mathematical tools from network science and graph theory to quantify the system’s topological organization and to better understand its function. Here, we used two-photon calcium imaging to record spontaneous activity from the same set of cells in mouse auditory cortex over the course of several weeks. We reconstruct functional networks in which cells are linked to one another by edges weighted according to the correlation of their fluorescence traces. We show that the networks exhibit modular structure across multiple topological scales and that these multi-scale modules unfold as part of a hierarchy. We also show that, on average, network architecture becomes increasingly dissimilar over time, with similarity decaying monotonically with the distance (in time) between sessions. Finally, we show that a small fraction of cells maintain strongly-correlated activity over multiple days, forming a stable temporal core surrounded by a fluctuating and variable periphery. Our work indicates a framework for studying spontaneous activity measured by two-photon calcium imaging using computational methods and graphical models from network science. The methods are flexible and easily extended to additional datasets, opening the possibility of studying cellular level network organization of neural systems and how that organization is modulated by stimuli or altered in models of disease.

Neurons coordinate their activity with one another, forming networks that help support adaptive, flexible behavior. Still, little is known about the organization of these networks at the cellular scale and their stability over time. Here, we reconstruct networks from calcium imaging data recorded in mouse primary auditory cortex. We show that these networks exhibit spatially constrained, hierarchical modular structure, which may facilitate specialized information processing. However, we show that connection weights and modular structure are also variable over time, changing on a timescale of days and adopting novel network configurations. Despite this, a small subset of neurons maintain their connections to one another and preserve their modular organization across time, forming a stable temporal core surrounded by a flexible periphery. These findings represent a conceptual bridge linking network analyses of macroscale and cellular-level neuroimaging data. They also represent a complementary approach to existing circuits- and systems-based interrogation of nervous system function, opening the door for deeper and more targeted analysis in the future.

Model genotype-phenotype mappings and the algorithmic structure of evolution

Cancers are complex dynamic systems that undergo evolution and selection. Personalized medicine approaches in the clinic increasingly rely on predictions of tumour response to one or more therapies; these predictions are complicated by the inevitable evolution of the tumour. Despite enormous amounts of data on the mutational status of cancers and numerous therapies developed in recent decades to target these mutations, many of these treatments fail after a time due to the development of resistance in the tumour. The emergence of these resistant phenotypes is not easily predicted from genomic data, since the relationship between genotypes and phenotypes, termed the genotype-phenotype (GP) mapping, is neither injective nor functional. We present a review of models of this mapping within a generalized evolutionary framework that takes into account the relation between genotype, phenotype, environment and fitness. Different modelling approaches are described and compared, and many evolutionary results are shown to be conserved across studies despite using different underlying model systems. In addition, several areas for future work that remain understudied are identified, including plasticity and bet-hedging. The GP-mapping provides a pathway for understanding the potential routes of evolution taken by cancers, which will be necessary knowledge for improving personalized therapies.

Cell signaling heterogeneity is modulated by both cell-intrinsic and -extrinsic mechanisms: An integrated approach to understanding targeted therapy

During the last decade, our understanding of cancer cell signaling networks has significantly improved, leading to the development of various targeted therapies that have elicited profound but, unfortunately, short-lived responses. This is, in part, due to the fact that these targeted therapies ignore context and average out heterogeneity. Here, we present a mathematical framework that addresses the impact of signaling heterogeneity on targeted therapy outcomes. We employ a simplified oncogenic rat sarcoma (RAS)-driven mitogen-activated protein kinase (MAPK) and phosphoinositide 3-kinase-protein kinase B (PI3K-AKT) signaling pathway in lung cancer as an experimental model system and develop a network model of the pathway. We measure how inhibition of the pathway modulates protein phosphorylation as well as cell viability under different microenvironmental conditions. Training the model on this data using Monte Carlo simulation results in a suite of in silico cells whose relative protein activities and cell viability match experimental observation. The calibrated model predicts distributional responses to kinase inhibitors and suggests drug resistance mechanisms that can be exploited in drug combination strategies. The suggested combination strategies are validated using in vitro experimental data. The validated in silico cells are further interrogated through an unsupervised clustering analysis and then integrated into a mathematical model of tumor growth in a homogeneous and resource-limited microenvironment. We assess posttreatment heterogeneity and predict vast differences across treatments with similar efficacy, further emphasizing that heterogeneity should modulate treatment strategies. The signaling model is also integrated into a hybrid cellular automata (HCA) model of tumor growth in a spatially heterogeneous microenvironment. As a proof of concept, we simulate tumor responses to targeted therapies in a spatially segregated tissue structure containing tumor and stroma (derived from patient tissue) and predict complex cell signaling responses that suggest a novel combination treatment strategy.

A signaling pathway is a network of molecules in a cell that is typically initiated by stimuli (e.g., microenvironmental cues) acting on receptors and internal signaling molecules to determine cell fate. Signaling pathways in cancer cells are different from those in normal cells, and this difference helps cancer cells to grow and thrive indefinitely. Drugs that target the aberrant signaling pathways in cancer cells (often referred to as targeted therapy) are promising for improving treatment outcomes of many different cancers in patients. However, most patients eventually develop resistance to these drugs. Resistance may already be present in the tumor or may emerge via mutation or via microenvironmental mediation. Tumor heterogeneity, which is characterized by subtle or dramatic differences among tumor cells, plays a key role in the development of drug resistance. Some tumor cells respond well to therapy, while others may adapt to the stress induced by the drug within the microenvironment. Moreover, removal of drug-sensitive cells may result in the competitive release of drug-resistant cells. Here, we present mathematical models to assess the impact of heterogeneity in signaling pathways within tumor cells on the outcomes of targeted therapy. We consider a simplified version of two well-known signaling pathways that modulate the growth of lung cancer cells. By using different targeted therapies, we quantify the effect of pathway inhibition on protein activity and cell viability and developed a mathematical model of the network, which is trained to reproduce these data and to develop a panel of heterogeneous in silico cells. The model predicts potential mechanisms of drug resistance and proposes combination therapies that are effective across the panel. We validate these combination therapies experimentally using the lung cancer cells and integrated the in silico cells into a computational lung tissue model that explicitly captures the microenvironment of lung cancer. Our results suggest that heterogeneity in both the tumor and microenvironment impacts treatment response in different ways and suggest a novel combination therapy for a better response.

Stochasticity in the Genotype-Phenotype Map: Implications for the Robustness and Persistence of Bet-Hedging

Nongenetic variation in phenotypes, or bet-hedging, has been observed as a driver of drug resistance in both bacterial infections and cancers. Here, we study how bet-hedging emerges in genotype-phenotype (GP) mapping through a simple interaction model: a molecular switch. We use simple chemical reaction networks to implement stochastic switches that map gene products to phenotypes, and investigate the impact of structurally distinct mappings on the evolution of phenotypic heterogeneity. Bet-hedging naturally emerges within this model, and is robust to evolutionary loss through mutations to both the expression of individual genes, and to the network itself. This robustness explains an apparent paradox of bet-hedging-why does it persist in environments where natural selection necessarily acts to remove it? The structure of the underlying molecular mechanism, itself subject to selection, can slow the evolutionary loss of bet-hedging to ensure a survival mechanism against environmental catastrophes even when they are rare. Critically, these properties, taken together, have profound implications for the use of treatment-holidays to combat bet-hedging-driven resistant disease, as the efficacy of breaks from treatment will ultimately be determined by the structure of the GP mapping.

The Importance of Neighborhood Scheme Selection in Agent-based Tumor Growth Modeling

Modeling tumor growth has proven a very challenging problem, mainly due to the fact that tumors are highly complex systems that involve dynamic interactions spanning multiple scales both in time and space. The desire to describe interactions in various scales has given rise to modeling approaches that use both continuous and discrete variables, known as hybrid approaches. This work refers to a hybrid model on a 2D square lattice focusing on cell movement dynamics as they play an important role in tumor morphology, invasion and metastasis and are considered as indicators for the stage of malignancy used for early prognosis and effective treatment. Considering various distributions of the microenvironment, we explore how Neumann vs. Moore neighborhood schemes affects tumor growth and morphology. The results indicate that the importance of neighborhood selection is critical under specific conditions that include i) increased hapto/chemo-tactic coefficient, ii) a rugged microenvironment and iii) ECM degradation.

Machine learning for biomarker identification in cancer research - developments toward its clinical application

The patterns identified from the systematically collected molecular profiles of patient tumor samples, along with clinical metadata, can assist personalized treatments for effective management of cancer patients with similar molecular subtypes. There is an unmet need to develop computational algorithms for cancer diagnosis, prognosis and therapeutics that can identify complex patterns and help in classifications based on plethora of emerging cancer research outcomes in public domain. Machine learning, a branch of artificial intelligence, holds a great potential for pattern recognition in cryptic cancer datasets, as evident from recent literature survey. In this review, we focus on the current status of machine learning applications in cancer research, highlighting trends and analyzing major achievements, roadblocks and challenges toward its implementation in clinics.