Perturbation biology links temporal protein changes to drug responses in a melanoma cell line

Understanding molecular signaling cascades in neural disease using multi-resolution models

If the genome defines the program for the operations of a cell, signaling networks execute it. These cascades of chemical, cell-biological, structural, and trafficking events span milliseconds (e.g., synaptic release) to potentially a lifetime (e.g., stabilization of dendritic spines). In principle almost every aspect of neuronal function, particularly at the synapse, depends on signaling. Thus dysfunction of these cascades, whether through mutations, local dysregulation, or infection, leads to disease. The sheer complexity of these pathways is matched by the range of diseases and the diversity of their phenotypes. In this review, we discuss how to build computational models, how these models are essential to tackle this complexity, and the benefits of using families of models at different levels of detail to understand signaling in health and disease.

A comprehensive mechanistic model of adipocyte signaling with layers of confidence

Adipocyte signaling, normally and in type 2 diabetes, is far from fully understood. We have earlier developed detailed dynamic mathematical models for several well-studied, partially overlapping, signaling pathways in adipocytes. Still, these models only cover a fraction of the total cellular response. For a broader coverage of the response, large-scale phosphoproteomic data and systems level knowledge on protein interactions are key. However, methods to combine detailed dynamic models with large-scale data, using information about the confidence of included interactions, are lacking. We have developed a method to first establish a core model by connecting existing models of adipocyte cellular signaling for: (1) lipolysis and fatty acid release, (2) glucose uptake, and (3) the release of adiponectin. Next, we use publicly available phosphoproteome data for the insulin response in adipocytes together with prior knowledge on protein interactions, to identify phosphosites downstream of the core model. In a parallel pairwise approach with low computation time, we test whether identified phosphosites can be added to the model. We iteratively collect accepted additions into layers and continue the search for phosphosites downstream of these added layers. For the first 30 layers with the highest confidence (311 added phosphosites), the model predicts independent data well (70-90% correct), and the predictive capability gradually decreases when we add layers of decreasing confidence. In total, 57 layers (3059 phosphosites) can be added to the model with predictive ability kept. Finally, our large-scale, layered model enables dynamic simulations of systems-wide alterations in adipocytes in type 2 diabetes.

Network inference from perturbation time course data

Networks underlie much of biology from subcellular to ecological scales. Yet, understanding what experimental data are needed and how to use them for unambiguously identifying the structure of even small networks remains a broad challenge. Here, we integrate a dynamic least squares framework into established modular response analysis (DL-MRA), that specifies sufficient experimental perturbation time course data to robustly infer arbitrary two and three node networks. DL-MRA considers important network properties that current methods often struggle to capture: (i) edge sign and directionality; (ii) cycles with feedback or feedforward loops including self-regulation; (iii) dynamic network behavior; (iv) edges external to the network; and (v) robust performance with experimental noise. We evaluate the performance of and the extent to which the approach applies to cell state transition networks, intracellular signaling networks, and gene regulatory networks. Although signaling networks are often an application of network reconstruction methods, the results suggest that only under quite restricted conditions can they be robustly inferred. For gene regulatory networks, the results suggest that incomplete knockdown is often more informative than full knockout perturbation, which may change experimental strategies for gene regulatory network reconstruction. Overall, the results give a rational basis to experimental data requirements for network reconstruction and can be applied to any such problem where perturbation time course experiments are possible.

Hypothesis-generating proteome perturbation to identify NEU-4438 and acoziborole modes of action in the African Trypanosome

NEU-4438 is a lead for the development of drugs against Trypanosoma brucei, which causes human African trypanosomiasis. Optimized with phenotypic screening, targets of NEU-4438 are unknown. Herein, we present a cell perturbome workflow that compares NEU-4438's molecular modes of action to those of SCYX-7158 (acoziborole). Following a 6 h perturbation of trypanosomes, NEU-4438 and acoziborole reduced steady-state amounts of 68 and 92 unique proteins, respectively. After analysis of proteomes, hypotheses formulated for modes of action were tested: Acoziborole and NEU-4438 have different modes of action. Whereas NEU-4438 prevented DNA biosynthesis and basal body maturation, acoziborole destabilized CPSF3 and other proteins, inhibited polypeptide translation, and reduced endocytosis of haptoglobin-hemoglobin. These data point to CPSF3-independent modes of action for acoziborole. In case of polypharmacology, the cell-perturbome workflow elucidates modes of action because it is target-agnostic. Finally, the workflow can be used in any cell that is amenable to proteomic and molecular biology experiments.

From single-omics to interactomics: How can ligand-induced perturbations modulate single-cell phenotypes?

Cells suffer from perturbations by different stimuli, which, consequently, rise to individual alterations in their profile and function that may end up affecting the tissue as a whole. This is no different if we consider the effect of a therapeutic agent on a biological system. As cells are exposed to external ligands their profile can change at different single-omics levels. Detecting how these changes take place through different sequencing technologies is key to a better understanding of the effects of therapeutic agents. Single-cell RNA-sequencing stands out as one of the most common approaches for cell profiling and perturbation analysis. As a result, single-cell transcriptomics data can be integrated with other omics data sources, such as proteomics and epigenomics data, to clarify the perturbation effects and mechanism at the cell level. Appropriate computational tools are key to process and integrate the available information. This chapter focuses on the recent advances on ligand-induced perturbation and single-cell omics computational tools and algorithms, their current limitations, and how the deluge of data can be used to improve the current process of drug research and development.

Combination treatment optimization using a pan-cancer pathway model

The design of efficient combination therapies is a difficult key challenge in the treatment of complex diseases such as cancers. The large heterogeneity of cancers and the large number of available drugs renders exhaustive in vivo or even in vitro investigation of possible treatments impractical. In recent years, sophisticated mechanistic, ordinary differential equation-based pathways models that can predict treatment responses at a molecular level have been developed. However, surprisingly little effort has been put into leveraging these models to find novel therapies. In this paper we use for the first time, to our knowledge, a large-scale state-of-the-art pan-cancer signaling pathway model to identify candidates for novel combination therapies to treat individual cancer cell lines from various tissues (e.g., minimizing proliferation while keeping dosage low to avoid adverse side effects) and populations of heterogeneous cancer cell lines (e.g., minimizing the maximum or average proliferation across the cell lines while keeping dosage low). We also show how our method can be used to optimize the drug combinations used in sequential treatment plans—that is, optimized sequences of potentially different drug combinations—providing additional benefits. In order to solve the treatment optimization problems, we combine the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) algorithm with a significantly more scalable sampling scheme for truncated Gaussian distributions, based on a Hamiltonian Monte-Carlo method. These optimization techniques are independent of the signaling pathway model, and can thus be adapted to find treatment candidates for other complex diseases than cancers as well, as long as a suitable predictive model is available.

Combination therapies are a promising approach to counter complex diseases such as cancers. Two key difficulties in the design of effective cancer combination therapies are the large number of available drugs and the heterogeneity of cancers which render exhaustive laboratory studies impractical. In recent years, sophisticated signaling pathway models that can predict responses to combination treatments at a molecular level have been developed. This motivates the question of how one can leverage mechanistic models to identify candidates for novel combination treatments. In this paper we propose a combination treatment optimization framework which employs a large-scale pan-cancer pathway model. We formulate treatment optimization problems for single cell lines and heterogeneous populations of cancer cells. We further investigate sequential treatment plans and combine an existing evolutionary algorithm with an efficient Hamiltonian Monte-Carlo based sampling scheme. During extensive simulation studies our approach identified combination therapies which are predicted to be more effective than conventional treatments. We hope that one day in silico experiments will be used to identify a small set of promising treatment candidates which can then form a starting point for laboratory studies, allowing for an efficient use of limited resources and accelerated discovery of effective therapies.

HillTau: A fast, compact abstraction for model reduction in biochemical signaling networks

Signaling networks mediate many aspects of cellular function. The conventional, mechanistically motivated approach to modeling such networks is through mass-action chemistry, which maps directly to biological entities and facilitates experimental tests and predictions. However such models are complex, need many parameters, and are computationally costly. Here we introduce the HillTau form for signaling models. HillTau retains the direct mapping to biological observables, but it uses far fewer parameters, and is 100 to over 1000 times faster than ODE-based methods. In the HillTau formalism, the steady-state concentration of signaling molecules is approximated by the Hill equation, and the dynamics by a time-course tau. We demonstrate its use in implementing several biochemical motifs, including association, inhibition, feedforward and feedback inhibition, bistability, oscillations, and a synaptic switch obeying the BCM rule. The major use-cases for HillTau are system abstraction, model reduction, scaffolds for data-driven optimization, and fast approximations to complex cellular signaling.

Chemical signals mediate many computations in cells, from housekeeping functions in all cells to memory and pattern selectivity in neurons. These signals form complex networks of interactions. Computer models are a powerful way to study how such networks behave, but it is hard to get all the chemical details for typical models, and it is slow to run them with standard numerical approaches to chemical kinetics. We introduce HillTau as a simplified way to model complex chemical networks. HillTau models condense multiple reaction steps into single steps defined by a small number of parameters for activation and settling time. As a result the models are simple, easy to find values for, and they run quickly. Remarkably, they fit the full chemical formulations rather well. We illustrate the utility of HillTau for modeling several signaling network functions, and for fitting complicated signaling networks.

PHENSIM: Phenotype Simulator

Despite the unprecedented growth in our understanding of cell biology, it still remains challenging to connect it to experimental data obtained with cells and tissues’ physiopathological status under precise circumstances. This knowledge gap often results in difficulties in designing validation experiments, which are usually labor-intensive, expensive to perform, and hard to interpret. Here we propose PHENSIM, a computational tool using a systems biology approach to simulate how cell phenotypes are affected by the activation/inhibition of one or multiple biomolecules, and it does so by exploiting signaling pathways. Our tool’s applications include predicting the outcome of drug administration, knockdown experiments, gene transduction, and exposure to exosomal cargo. Importantly, PHENSIM enables the user to make inferences on well-defined cell lines and includes pathway maps from three different model organisms. To assess our approach’s reliability, we built a benchmark from transcriptomics data gathered from NCBI GEO and performed four case studies on known biological experiments. Our results show high prediction accuracy, thus highlighting the capabilities of this methodology. PHENSIM standalone Java application is available at https://github.com/alaimos/phensim, along with all data and source codes for benchmarking. A web-based user interface is accessible at https://phensim.tech/.

Despite the unprecedented growth in our understanding of cell biology, it still remains challenging to connect it to experimental data obtained with cells and tissues’ physiopathological status under precise circumstances. This knowledge gap often results in difficulties in designing validation experiments, which are usually labor-intensive, expensive to perform, and hard to interpret. In this context, ’in silico’ simulations can be extensively applied in massive scales, testing thousands of hypotheses under various conditions, which is usually experimentally infeasible. At present, many simulation models have become available. However, complex biological networks might pose challenges to their performance.

We propose PHENSIM, a computational tool using a systems biology approach to simulate how cell phenotypes are affected by the activation/inhibition of one or multiple biomolecules, and it does so by exploiting signaling pathways. We implemented our tool as a freely accessible web application, hoping to allow ’in silico’ simulations to play a more central role in the modeling and understanding of biological phenomena.

CellBox: Interpretable Machine Learning for Perturbation Biology with Application to the Design of Cancer Combination Therapy

Systematic perturbation of cells followed by comprehensive measurements of molecular and phenotypic responses provides informative data resources for constructing computational models of cell biology. Models that generalize well beyond training data can be used to identify combinatorial perturbations of potential therapeutic interest. Major challenges for machine learning on large biological datasets are to find global optima in a complex multidimensional space and mechanistically interpret the solutions. To address these challenges, we introduce a hybrid approach that combines explicit mathematical models of cell dynamics with a machine-learning framework, implemented in TensorFlow. We tested the modeling framework on a perturbation-response dataset of a melanoma cell line after drug treatments. The models can be efficiently trained to describe cellular behavior accurately. Even though completely data driven and independent of prior knowledge, the resulting de novo network models recapitulate some known interactions. The approach is readily applicable to various kinetic models of cell biology. A record of this paper's Transparent Peer Review process is included in the Supplemental Information.