Deep-learning-based Estimation of Heterogeneous persistence length of Bio-filaments

Abstract

Cellular bio-filaments often exhibit compositional heterogeneities due to various factors, including the binding of associated proteins, the formation of structural defects during polymerization, or the assembly of different types of monomers. These heterogeneities result in spatially varying mechanical and kinetic properties of the bio-filaments, and understanding how local effects augment the material properties of the bio-filaments in a spatially resolved manner is essential for comprehending how binding and bundling regulate bio-filaments functions. In this study, we develop a deep-learning-based method designed to estimate heterogeneous persistence length profiles of bio-filaments as a function of the arc-length. To validate our methodology, we utilize computer-generated bio-filaments with known persistence length profiles and compare its performance to the conventional methodology that employs the tangent-tangent correlation function in various scenarios. Our results demonstrate that our methodology performs comparably or even better than the conventional approach. Additionally, we apply our method to an experimental image of B-DNA, yielding a homogeneous persistence length of 42.5 nm, consistent with the well-known value of 45 nm.