Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Thesis_Shishir.pdf (8.36 MB)

ETD Abstract Container

Abstract Header

STATISTICAL PHYSICS OF CELL ADHESION COMPLEXES AND MACHINE LEARNING

Adhikari, Shishir Raj
http://rave.ohiolink.edu/etdc/view?acc_num=case1562167640484477

Abstract Details

2019, Doctor of Philosophy, Case Western Reserve University, Physics.
In the last 20 years, there have been huge advances in the field of single-molecule force spectroscopy experimental techniques. With such advances, there is a plethora of raw single-molecule data. In those data, one of the most intriguing observations is the existence of biphasic bond lifetime behavior (catch bonds) in the protein-ligand interaction system under the application of the force. The first part of the thesis focuses on a theoretical way of understanding the origin of catch bonds in the cadherin-catenin-actin (CCA) and L-selectin-ligand systems. In both cases, we show that only a mode with two degrees of freedom is sufficient to reproduce catch-bond behavior. We also fit our model to the experimental bond lifetime data and learn that the value of the free parameters extracted from fitting corroborates with observed values from the protein structure. Furthermore, we also explore the non-Markovian behavior observed in the L-selectin-ligand system under the application of force at different ramping rates. In the case of varying the ramping rate, we learn that ramping behavior induces changes to the protein-ligand interface analogous to introducing mutations in the lectin domain of the L-selectin protein. In the second part of the thesis, we present the mapping between machine learning dynamics and non-equilibrium statistical mechanics. We focus on stochastic gradient descent (SGD) learning dynamics and map SGD to the Fokker-Planck dynamics. Using Fokker-Planck dynamics, we characterize the steady state probability distribution of the weights (the parameters of the machine learning algorithm). We learn that steady state probability distribution is non-Boltzmannian, which means that the SGD dynamics leads to a non-equilibrium steady state. By forcing thermalization, we also get a notion of temperature for the ML system and a weight update rule similar to natural gradient descent.
Michael Hinczewski (Committee Chair)
Philip Taylor (Committee Member)
Lydia Kisley (Committee Member)
Alkan Kabakcioglu (Committee Member)
Steven Izen (Committee Member)
152 p.
Biophysics; Physics
Catch bond, protein-ligand interaction, machine learning, stochastic gradient descent, Fokker-Planck equation, L-selectin, Cadherin, Catenin, Actin, Triphasic bond, Slip bond, bond dynamics modeling, First passage time

Recommended Citations

Citations

  • Adhikari, S. R. (2019). STATISTICAL PHYSICS OF CELL ADHESION COMPLEXES AND MACHINE LEARNING [Doctoral dissertation, Case Western Reserve University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=case1562167640484477

    APA Style (7th edition)

  • Adhikari, Shishir. STATISTICAL PHYSICS OF CELL ADHESION COMPLEXES AND MACHINE LEARNING. 2019. Case Western Reserve University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=case1562167640484477.

    MLA Style (8th edition)

  • Adhikari, Shishir. "STATISTICAL PHYSICS OF CELL ADHESION COMPLEXES AND MACHINE LEARNING." Doctoral dissertation, Case Western Reserve University, 2019. http://rave.ohiolink.edu/etdc/view?acc_num=case1562167640484477

    Chicago Manual of Style (17th edition)

case1562167640484477
542
© 2019, all rights reserved.
This open access ETD is published by Case Western Reserve University School of Graduate Studies and OhioLINK.