Inferring Kinetic Models for Large-Scale Biochemical Networks
EMSL Project ID
60374
Abstract
Mechanistic models of cell networks that accurately predict changes in input-output dynamics and outcomes upon perturbation have the potential to facilitate a greater understanding of the interactions within signaling and metabolic networks and allow for the manipulation of pathways for beneficial effects like drug discovery and metabolic engineering. In addition to traditional approaches that construct literature-based networks and calibrate them to data, recent work has focused on generating mechanistic models in a purely data-driven approach like the sparse identification of nonlinear dynamics (SINDy). However, cell networks are complex with numerous interactions and many overlapping pathways. There are potentially many unknowns and simplifications must often be made for the sake of tractability. These issues along with data that is often paltry compared to the size of the model renders the construction of predictive models via either method difficult. To address the problem of predictability in mechanistic models, we have developed software for the rapid creation of synthetic kinetic models that can mimic the properties of real networks and can act as benchmarks for testing of model inference algorithms. In this proposal a set of benchmarks and associated synthetic data will be generated. The data will be used in a cross-validation formalism to evaluate the predictive power of the models generated via variants of the SINDy and similar algorithms. We will determine the quantity and positions of data necessary to reproduce the benchmarks. We will also develop an iterative model ensemble approach for model building that incorporates existing knowledge and is designed explicitly for hypothesis generation. Adversarial machine learning will be used to evaluate the ability of this method to recapitulate the benchmark models. This endeavor will require significant computational resources but will potentially have a significant impact on the development of more predictive mechanistic models.
Project Details
Project type
Large-Scale EMSL Research
Start Date
2022-10-01
End Date
N/A
Status
Active
Released Data Link
Team
Principal Investigator
Team Members