Mechanistic mathematical models of biological systems have been used to describe biological phenomena, including human disease, in the hope that one day these models might be used to hypothesize and optimize novel treatment strategies. Current evolutionary algorithmic approaches can be used to symbolically regress mathematical models from data; however, they are restricted to the formulation of simple or approximate models due to the computational cost of evolving mechanistic models for complex systems. It was hypothesized that bounds on the problem's solution space could be implemented using reaction kinetic theory and thus shrink the solution space needing to be searched.  Using such an approach, inference of mechanistic mathematical models consisting of complex biological interactions could potentially become tractable. A step-by-step approach was developed involving breaking up the original problem into smaller problems and focusing on the reaction sub-network of one chemical species at a time. The sub-networks could then be re-integrated allowing build up to a complete solution for the entire reaction network. The methodology was evaluated by creating an in silico data set using a three-component non-linear viral dynamic model and attempting to infer the original reaction kinetic network.  Without any noise added to the data, it was indeed possible to regenerate the original model. When a noisy in silico data set with 10% Gaussian noise was utilized, two solutions with identical fitness scores, according to the Akaike Information Criterion with correction, were found. An additional in silico data set with 10% Gaussian noise was used to identify one of these models as optimal. This optimized solution was found to be the same as the original general viral dynamics model.

Discovery of biochemical kinetics and mechanism using evolutionary algorithms

247th National Meeting (2014)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.