Predictive Modeling for Geometric Rule-Based Methods

Previously, we proposed a method for incorporating molecular geometry in a biological rule-based model by encoding molecular curvature into the rules and associated binding rate constants. We combined this method with a 3D rigid-body Monte Carlo simulation to model antigen-antibody aggregation. In this work, we use our geometric rule-based method to develop a model for predicting the output of the full-resolution Monte Carlo simulation given the output of lower resolution simulations. The purpose of this predictive model is to reduce the computational cost of the Monte Carlo simulation. We develop this model by first choosing a rule set for each molecular geometry and varying only the binding rate constant for each Monte Carlo resolution, and then fitting the resulting data to a function. We examine the calculation time needed for each predictive model to demonstrate how this model is more efficient than running a full-resolution simulation. We find that this method can reduce the computational time of the Monte Carlo simulation by up to 20%.