Inferring bending persistence length of biofilaments using a neural network method

Biofilaments, including DNA, actin, and microtubules, often display spatial variations in bending stiffness due to influences such as internal structural heterogeneities, protein interactions, and environmental influences in intracellular settings. Accurately mapping these variations is crucial for understanding biofilament behaviors, but is complicated by the limited experimental data available. In this study, we introduce a novel approach based on a neural network for estimating spatially varying bending stiffness in biofilaments. Leveraging a neural network model in conjunction with a sliding window technique, our method successfully extracts stiffness patterns from polymer configurations. Through tests on simulated data, we show that it surpasses conventional approaches based on the tangent-tangent correlation functions, especially under conditions with sparse data. Our model also uncovers features akin to turning angles to infer stiffness properties. Furthermore, we apply the method to both experimental and simulated AFM images, producing estimates in line with known persistence length values. This framework provides a robust tool for investigating the mechanical properties of biofilaments and holds promise for broader applications in experimental studies.