Automated analysis of AFM images of membrane proteins via conventional and deep learning methods
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
This dissertation develops computational approaches to automate the analysis of membraneassociated proteins imaged by atomic force microscopy (AFM). Cellular membranes interact with numerous proteins whose folding, assembly, and conformational dynamics are essential to cellular function. Here, we focus on two systems: Candidalysin (CL), the poreforming virulence factor of Candida albicans, and SecYEG, a core protein in the general secretory system of E. coli. We introduce PsPolypy, an open-source Python toolkit that detects polymeric particles in AFM images and automatically performs skeletonization, topology classification, and persistence length calculation to quantify Candidalysin polymerization and bending mechanics. We then compare traditional image classification methods with convolutional neural networks for supervised classification of the protruding sides of SecYEG. Finally, we develop an unsupervised deep learning workflow that clusters protein conformations and applies localization atomic force microscopy to enhance in-plane AFM resolution, validated on synthetic AFM images generated from all-atom molecular dynamics simulations. Together, these studies establish a framework that bridges experimental and computational techniques to quantitatively characterize protein structure and dynamics from AFM data.
Table of Contents
PubMed ID
Degree
Ph. D.