With nanometer-level precision, virus infection can be detected within 1 hour. Southern Medical University's cell nucleus AI tool is published in Nature sub-journal

Editor | KX

A nanometer (nm) is one billionth of a meter, and the width of a human hair is about 100,000 nm.

Today, artificial intelligence can detect rearrangements within cells as small as 20 nm, or 5,000 times smaller than the width of a human hair. These changes are too small and subtle to be discovered by humans using traditional methods alone.

Recently, a research team from Southern Medical University and the Barcelona Institute of Science and Technology in Spain developed a nuclear artificial intelligence (AINU) tool that can identify specific nuclear features at nanometer-level resolution. It can distinguish cancer cells from normal cells and detect early stages of intracellular viral infection.

Limei Zhong, co-author of the paper and researcher at Guangdong Provincial People's Hospital (GDPH) of Southern Medical University, said: "Researchers can use this technology to observe how the virus affects cells immediately after entering the human body, which can help develop better treatments and vaccines. In hospitals and clinics, AINU can be used to quickly diagnose infections from simple blood or tissue samples, making the diagnostic process faster and more accurate. "Related research is based on "A deep learning method that "Identifies cellular heterogeneity using nanoscale nuclear features" was published in "Nature Machine Intelligence".

Paper link: https://www.nature.com/articles/s42256-024-00883-x Nanoscale resolution microscopy

Cellular phenotypic heterogeneity is many Key determinants of biological function, understanding their origins remains a formidable challenge. This heterogeneity often reflects changes in chromatin structure, influenced by factors such as viral infection and cancer, which dramatically reshape the cellular landscape.

Single molecule localization microscopy (SMLM), specifically stochastic optical reconstruction microscopy (STORM), can determine the nanoscale arrangement of chromatin fibers in cells. Current methods for analyzing the spatial distribution of single molecules, such as clustering algorithms, are very effective in extracting nuclear positions and their local densities. However, it is currently unclear how the spatial distribution and density of these molecules can be exploited to identify cellular states.

Convolutional neural networks (CNN) have been widely used in various healthcare imaging fields. Deep learning (DL) models have been used to classify whole-cell images and track them using diffraction-limited microscopy. Additionally, super-resolution (SR) microscopy is used to improve localization accuracy and semantic segmentation during data acquisition, but SMLM images have not yet been used to classify cells based on subcellular structure.

Molecular-level "facial recognition"

Unlocking smartphones with your face, or self-driving cars understanding and navigating the environment by identifying objects on the road, all make use of convolutional neural networks.

In the medical field, convolutional neural networks are used to analyze medical images, such as mammograms or CT scans, and identify signs of cancer that the human eye might miss. They can also help doctors detect abnormalities in MRI scans or X-ray images, helping doctors make diagnoses faster and more accurately.

AINU is a convolutional neural network, a type of AI specifically designed to analyze visual data such as images. CNN architectures can be efficiently trained using minimal training data from nuclear signature imaging.

AINU scans high-resolution images of cells, which are obtained with STORM, a technology that captures finer details than ordinary microscopes. High-definition snapshots can reveal structures with nanometer-scale resolution.

"The resolution of these images is high enough for our AI to identify specific patterns and differences, including changes in the arrangement of DNA within cells, with astonishing accuracy, helping us detect changes very quickly. I We believe that this kind of information could one day buy doctors valuable time to monitor disease, personalize treatments and improve patient outcomes," said study co-corresponding author Professor Pia Cosma from the Institute of Science and Technology in Barcelona, ​​Spain.

To select the best CNN architecture and its hyperparameters for identifying somatic cells and human induced pluripotent stem cells (hiPSCs), the researchers compared 11 different CNN architectures, and finally, DenseNet-121 performed better in identifying somatic cells and human induced pluripotent stem cells (hiPSCs). Somatic cells and hiPSCs performed best, with an average validation accuracy of 92.26 and an average loss of 0.292, which were used for subsequent analysis.

Selection was based on model performance on a total of 349 nuclear two-color STORM images of nucleosome core histones H3 and Pol II. Fluorophores of selected molecules were collected from human somatic cells and hiPSCs of different somatic cell types and rendered into images at 10x magnification relative to the original camera frame.

AINU detects and analyzes tiny structures within cells at the molecular level. The researchers trained the model by feeding it nanometer-resolution images of the nuclei of different types of cells in different states. The model learned to recognize specific patterns in cells by analyzing how components of the nucleus are distributed and arranged in three-dimensional space.

Zum Beispiel weisen Krebszellen im Vergleich zu normalen Zellen offensichtliche Veränderungen in ihrer Kernstruktur auf, etwa Veränderungen in der Art und Weise, wie ihre DNA organisiert ist oder in der Verteilung von Enzymen im Kern. Nach dem Training kann AINU neue Bilder von Zellkernen analysieren und sie allein aufgrund dieser Merkmale als Krebszellen oder normale Zellen klassifizieren.

Abbildung: Mit Pol II- und H3-Bildern trainiertes AINU identifiziert somatische Zellen und iPSCs korrekt. (Quelle: Paper) AINU ist in der Lage, verschiedene zelluläre Zustände basierend auf der räumlichen Anordnung von Kernhiston H3, RNA-Polymerase II (Pol II) oder DNA in hochauflösenden Mikroskopiebildern zu unterscheiden. Mit nur einer kleinen Anzahl von Bildern als Trainingsdaten kann AINU bei entsprechender Umschulung menschliche Körperzellen, vom Menschen induzierte pluripotente Stammzellen (iPSCs), mit dem Herpes-simplex-Virus Typ I (HSV-1) infizierte menschliche Zellen und Krebs genau identifizieren Zellen.

AINU identifiziert somatische Zellen und iPSCs

Abbildung: Auf Pol II-Bildern trainiertes AINU identifiziert somatische Zellen und iPSCs korrekt. (Quelle: Papier)

Aufdeckung von Identifikationsmerkmalen

Interpretierbare KI zeigt, dass die Lokalisierung von Pol II im Nukleolus ein Schlüsselmerkmal der AINU-Identifizierung von hiPSCs ist.

Erkennung von HSV-1

Die nanoskalige Auflösung der Bilder ermöglicht es der KI, Veränderungen im Zellkern innerhalb einer Stunde nach der Infektion der Zellen mit HSV-1 zu erkennen . Das Modell kann das Vorhandensein des Virus erkennen, indem es subtile Unterschiede in der Dichte der DNA findet.

Klinische Anwendungen

Forscher überwinden Einschränkungen bei der Verwendung dieser Technologie im klinischen Umfeld.

Beschleunigen Sie die wissenschaftliche Forschung

AINU identifiziert Stammzellen genau und hilft, die Stammzellenforschung zu beschleunigen.

Erkennung pluripotenter Zellen

AINU kann pluripotente Zellen schneller und genauer erkennen und so dazu beitragen, Stammzelltherapien sicherer und effektiver zu machen.

Reduzierung des Einsatzes von Tieren

Der Einsatz von AINU kann den Einsatz von Tieren in der Wissenschaft reduzieren.

Verwandte Berichte:

• https://medicalxpress.com/news/2024-08-ai-cancer-viral-infections-nanoscale.html

The above is the detailed content of With nanometer-level precision, virus infection can be detected within 1 hour. Southern Medical University's cell nucleus AI tool is published in Nature sub-journal. For more information, please follow other related articles on the PHP Chinese website!