Skip to main content

Machine learning and network analysis for macromolecular structure determination from super-resolution microscopy

Resource type
Thesis type
(Thesis) Ph.D.
Date created
2019-09-09
Authors/Contributors
Abstract
Single molecule localization microscopy (SMLM) is one of the super-resolution imaging techniques that break the diffraction limit barrier of light microscopy. SMLM generates a 3D point cloud from photon blinks generated by fluorophores binding to target proteins. With SMLM it is now possible to image subcellular structures and study their molecular organization. Following a review of state-of-the-art SMLM cluster analysis and quantification methods, we make the following computational contributions, which are applied to studying caveolin-1 (Cav1) protein biological structures (or Cav1 domains) in prostate cancer cells. (i) We propose the first graph-based machine learning pipeline to denoise, segment, and characterize Cav1 clusters (blobs) from SMLM point clouds. Our pipeline comprises computational modules to extract per-protein and per-blob features necessary to perform the aforementioned computational tasks. Our pipeline has enabled us to identify four distinct Cav1 domains, including caveolae as well as three smaller scaffold domains. (ii) We then propose a multi-proximity threshold graphlet analysis of Cav1 blobs to identify biosignatures of Cav1 domains (i.e. classify them into categories). Our graphlet analysis has resulted in discovering the patterns of molecular interactions in the four Cav1 domains, which defines the changes in the structural organization in caveolae and scaffolds. (iii) Subsequently, we compare the performance and tradeoffs between machine learning approaches (including deep learning), on different SMLM data representations. Our observation from this comparison is that deep learning models could be used to automatically learn biological features from 3D point cloud SMLM data that distinguish caveolae and non-caveolae structures. (iv) Finally, we leverage spectral network decomposition to detect modules within the various Cav1 domains at multiple proximity thresholds. By matching the biosignature across different Cav1 domains and their constituent modules, we decipher the assembly of complex structures and describe the biogenesis and formation of the caveolae in the membrane of endogenous HeLa cells. We conclude the thesis by discussing the potential applicability of the proposed methods to studying other proteins and biological structure, stating the limitations of the proposed methods, and providing our vision for future research directions.
Document
Identifier
etd20531
Copyright statement
Copyright is held by the author(s).
Permissions
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Supervisor or Senior Supervisor
Thesis advisor: Hamarneh, Ghassan
Language
English
Member of collection
Download file Size
etd20531_KhaterIsmail_revised.pdf 17.13 MB

Views & downloads - as of June 2023

Views: 33
Downloads: 1