Reconstruction and Modeling of Dynamical Molecular Networks: Abstract Biological networks and their quantitative models can provide mechanistic insights into pathophysiology of diseases as well as identify potential targets for therapeutic intervention. The quantitative models can be used for hypotheses generation through simulation of perturbations of key molecules and tested experimentally through pharmacological or genetic perturbations. This project deals with the development and implementation of algorithms and methodologies for causal inference, analysis and modeling of molecular and modular networks from large-scale temporal molecular data incorporating a priori knowledge related to biological pathways and functions. The dynamical and nonlinear nature of biological systems will be captured through successive linear models by identifying different temporal regimes in the time-course data. The temporal regimes will be identified through a change-point detection algorithm. The change-points potentially reflect mechanistic changes in the biological system. Then, a stable least absolute shrinkage and selection operator approach incorporating partial least squares will be used to infer the potentially causal networks and develop models for specific pathways. We will incorporate time-delay in our state-space modeling approach to identify if the data from the past contributes significantly to prediction of the current value. Since both inference and interpretation of large (causal) molecular networks from temporal data at the whole-systems level with thousands of components/molecules is prohibitively challenging, modules corresponding to various biological pathways, mechanisms and functions will be identified by integrating the quantitative temporal data and a priori biological knowledge. The hub-molecules or centroids of the modules will serve as state-variables in a reduced-dimensional state-space and they will be used to infer the networks and develop state-space models. The temporal evolution of the networks across various regimes will be rigorously analyzed by performing both qualitative and quantitative comparisons of the networks. The modular networks will also be compared with the corresponding coarse-grained versions of the detailed molecular networks as internal validation. External validations will include comparison with existing mechanistic models, if any. The predictive models of the networks will be used to generate experimentally testable hypotheses regarding temporally specific pharmacological perturbations of key proteins. While these methodologies will be applicable for many biological systems, in this project they will be applied to two systems, viz., 1) cell-cycle progression in mouse embryonic fibroblasts, important for the study of molecular mechanisms of cancer, and 2) differentiation of human induced pluripotent stem cells into neurons, important for the study of neurodegenerative diseases. The methods will be applied to simulated data as well. Statistical tools such as R and python will be used to implement the algorithms and methods and the resulting packages and tutorials will be made available to the research community through a PHP-based project website and public repositories such as GitHub and SourceForge.
Reconstruction and Modeling of Dynamical Molecular Networks: Narrative The major aim of this proposal is the development and implementation of algorithms and methodologies for causal inference, analysis and modeling of molecular and modular networks from large-scale temporal molecular data. The dynamical and nonlinear behavior of biological systems will be modeled through successive linear models incorporating time-delay using a stable least absolute shrinkage and selection operator approach for causal inference of networks, which, after reduction to a modular description, will also be used for developing state space models. While these methodologies will be widely applicable, two examples which span distinct biology, viz., cell-cycle progression in mouse embryonic fibroblasts, important for mechanistic understanding of cancer, and differentiation of human induced pluripotent stem cells into neurons important for neurodegenerative diseases, will be used to test, validate and exemplify the methods developed.