Characterizing the 1D diffusion dynamics of CRISPR-Cas9 complex via unsupervised machine learning

Abstract

CRISPR-Cas9 system is an adaptive immune system of prokaryotic organisms that recognizes and cleaves foreign genetic elements using RNA guide sequences. Due to this mechanism, the CRISPR-Cas9 system is widely used as a gene editing tool, but how this system searches for its target sequences is still elusive. Here, we design an experiment to observe the 1D diffusion of the CRISPR-Cas9 complex with 3 different seed sequences on a long DNA and employ an unsupervised machine learning framework to characterize the dynamics. In this framework, we first measure classical statistical quantities from all the trajectories and find that the dynamics of the CRISPR-Cas9 complex changes after a certain time. To characterize the dynamics, we define a vector extracting key features of each trajectory, visualize all the vectors in the 2D t-SNE space and perform clustering. As a result, we observe multiple distinct clusters in the 2D t-SNE space and find that each cluster shows different dynamics.