Related Experiment Video
Updated: Jan 12, 2026
12:06
Analyzing Mitochondrial Morphology Through Simulation Supervised Learning
Published on: March 3, 2023
4.6K
Label-Free Detection of Nuclear Envelope Nucleoporation using 2D Morphological Embeddings and Machine Learning
Keivan Rahmani1, Hamed Naghsh-Nilchi1, Leah Sadr1
1Aiiso Yufeng Li Family Department of Chemical and Nano Engineering, University of California, San Diego, CA, 92093, USA.
Small (Weinheim an Der Bergstrasse, Germany)
|November 7, 2025
Summary
This study introduces a machine learning method to detect nuclear envelope (NE) poration by analyzing cell and nuclear shape changes. This AI approach enables efficient, label-free monitoring of NE disruption for improved nuclear delivery applications.
Area of Science:
- Biotechnology
- Cell Biology
- Artificial Intelligence
Background:
- High-aspect-ratio nanostructures facilitate nuclear delivery via transient nuclear envelope (NE) disruption.
- Sporadic nucleoporation events limit the efficiency of current nuclear delivery methods.
Purpose of the Study:
- To develop a label-free, machine learning (ML) approach for detecting nucleoporation based on morphological cell and nuclear changes.
- To establish a high-throughput, non-invasive method for monitoring NE disruption events.
Main Methods:
- U2OS cells on silicon nanopillars were analyzed for NE disruption using Ku-80 mislocalization.
- A custom algorithm quantified Ku-80/DAPI intensity profiles to establish ground truth for cell states (intact vs. porated).
- An orientation-invariant variational autoencoder and support vector machine (SVM) were trained using cell/nuclear shape embeddings and morphological descriptors.
Main Results:
- The ML model achieved an 87.0% area under the receiver operating characteristic curve and 82.9% test accuracy.
- SHAP analysis identified nucleus-to-cell area ratio as the strongest predictor of nucleoporation.
- Specific nuclear features (e.g., surface smoothness, bulging) and cell features (e.g., boundary complexity, elongation) significantly influenced nucleoporation probability.
Conclusions:
- This AI workflow demonstrates a correlation between cell/nuclear morphology and nucleoporation events.
- The developed method enables non-invasive, high-throughput monitoring of cellular events using relatively small datasets.
- This approach holds potential for optimizing nuclear delivery strategies and investigating cellular phenotypes.
