Characterization of the Mechanism Behind Microtubule Dynamic Instability Using Artificial Inteligence

A microtubule (MT) is a cytoskeletal polymer that exhibits a behavior, known as dynamic instability (DI), where it switches randomly between phases of persistent growth and rapid shortening. The classical description of DI is based on the presence or absence of a GTP-cap that promotes polymerization. However, there is evidence that additional DI phases, like stutter, exist, suggesting this description needs to be updated. This work aims to elucidate the dimer-scale mechanisms behind DI by using Artificial Neural Networks (ANNs) and a previously validated dimer-scale model of MT DI. We report the development of ANNs capable of predicting the DI phase (i.e., growth, shortening, or stutter) of a simulated MT based only on its tip structure. These ANNs achieved greater than 80% accuracy on our test set and could achieve greater than 60% accuracy for examples of stutter. These results suggest that our ANNs identified features in MT tips that correlate with observed DI behavior and reinforce the existence of an additional DI phase, known as stutter. In ongoing work, we will use these ANNs to aid in identifying features that distinguish stutter from growth. Preliminary results, examining tip structures for which these ANNs are highly confident in their prediction, suggest that the presence of GDP subunits at the end of filaments is important to this distinction. This information will be used to formulate hypotheses of the dimer-scale mechanisms behind catastrophe.