Session: 07-02-01: Vibration and Acoustics in Biomedical Applications

Paper Number: 167225

Predicting Vibration-Induced Membrane Injury in HeLa Cells via Integrated Finite Element and Machine Learning Modeling 

Abstract

Introduction: Cells in living organisms are constantly subjected to mechanical stimuli that drive processes like differentiation, tissue repair, and injury response. While the cell membrane safeguards cellular integrity by regulating nutrient, waste, and signal flow, exposure to mechanical stress, from everyday forces to traumatic impacts, can deform the membrane and compromise its barrier function, ultimately resulting in increased permeability, disrupted signaling, inflammatory responses, or even cell death. Vibration is a unique type of mechanical stress that can affect cell membranes even at low, repeated levels. Over time, these vibrations may lead to subtle structural alterations that compromise membrane integrity. Investigating the impact of vibrational forces on cell membranes is critical not only for evaluating everyday and occupational exposures but also for guiding the design of safer equipment and more effective therapeutic strategies. This study focuses on predicting membrane injury in HeLa cells subjected to acoustic pressure-induced vibrations. Building on our previous work, where we developed a 2D finite element (FE) model in COMSOL Multiphysics, which accurately captured membrane deformation and cytosolic mechanics under acoustic pressure, we now integrate advanced machine learning (ML) algorithms to forecast both the onset and severity of membrane injury, thereby enhancing our ability to predict cellular responses across a wide range of vibrational conditions.

Methodology: Our FE model simulates a HeLa cell membrane under vibrational forces generated by acoustic pressure. The model incorporates critical input parameters, including membrane thickness, material properties of the cell interior and surrounding medium, pressure difference, acoustic pressure amplitude and frequency, time, and temperature. The simulation outputs include vertical and radial displacements of the membrane, areas of high stress, deformation patterns, strain distribution, pressure fields, cytosolic flow velocity, frequency response, energy absorption, and the temporal evolution of membrane deformation. Data from each simulation run are meticulously recorded in an Excel file, with each column representing a specific parameter, ensuring a comprehensive dataset for subsequent analysis.

The FE simulation data serves as the training set for various ML architectures developed to predict key stress responses, specifically von Mises and shear stresses. Our ML framework encompasses Multilayer Perceptron (MLP), 1D Convolutional Neural Networks (CNN), hybrid CNN-Recurrent Neural Networks (RNN), and autoencoder models with regression heads. Data preprocessing steps, such as standardization, missing value imputation, and train-test splitting, are carefully performed to ensure robustness and accuracy in the predictive models. This integrated computational framework builds a predictive model to effectively assess membrane injury risk under vibrational loads.

Results: The FE simulations consistently identified areas of high stress on the membrane and capture the dynamic interplay of force, displacement, and energy absorption. Analysis of the ML predictive models indicated that displacement is a primary factor influencing both shear and Von Mises stress distributions, key indicators of membrane integrity. Early results show promising accuracy in forecasting cellular responses across various acoustic scenarios, suggesting that the integration of FE models with ML successfully captures the complex interplay of forces involved in vibrational injury.

Conclusions: This study demonstrates that integrating FE models with advanced machine learning techniques provide a robust framework for predicting membrane injury due to vibrational forces. Our calibrated FE simulations replicate key experimental observations and accurately identify stress zones linked to potential injury. By elucidating the relationship between displacement and stress distribution, this approach offers valuable insights into the mechanical thresholds that trigger cellular injury. These findings are significant for developing safer devices and strategies to mitigate negative effects of mechanical vibrations. Future work will focus on refining ML algorithms and expanding simulation conditions to include a broader range of vibrational frequencies and amplitudes. Ultimately, this integrated computational approach paves the way for improved injury prevention strategies and innovative therapeutic interventions in cellular mechanics.

Presenting Author: Raheleh Miralami CAVS - Mississippi State University

Presenting Author Biography: Raheleh (Rahel) Miralami is an Assistant Research Professor at the Center for Advanced Vehicular Systems at Mississippi State University. With a B.S. and M.S. in Mechanical Engineering and a Ph.D. in Biomedical Science, she brings a robust interdisciplinary foundation to both her research and teaching. Her work spans biomechanics, mechanobiology, biomaterials, and bio-inspired design, seamlessly integrating computational modeling with experimental biomechanics across multiple scales.

Authors:

Eric Flynn Mabowitz Mississippi State University
Ethan Hendry Mississippi State University
Hamed Bakhtiarydavijani Mississippi State University
Noorbakhsh Amiri Golilarz The University of Alabama
Raheleh Miralami CAVS - Mississippi State University