(G-276) Protein recruitment detection after laser ablation inside nucleus using Machine Learning Algorithms

Introduction:: DNA double-stand breaks are the major cause of cancer-associated genome instability. Unrepaired DSBs can lead to chromosomal breakage and cell death, while incorrectly repaired DSBs would lead to genome instability. Our research interest is to understand the molecular mechanisms underlying the function of DNA damage checkpoint and DSB repair to maintain genome stability and prevent cancer and aging. We have established novel EGFP-based DSB repair assays and chromosomal translocation mouse models to investigate the mechanisms underlying DSB repair and chromosomal rearrangement. In the past 15 years, we applied the robotic laser microscope system to generate a precise damage line inside the nucleus of a U2OS cell to understand the genome-stability maintenance genes, such as Nbs1, Mre11, CtIP, Rad52 in the maintenance of genome stability. The GFP images before and after the laser ablation of the interested protein were acquired during the experiment. The recruitment line usually appeared right away, or a few minutes after the laser ablation. We used to spend tremendous time manually measuring the intensity of the recruitment line with respect to the time using ImageJ. To speed up the image processing, we proposed a line detecting algorithm using OpenCV2, and a pre-trained machine learning model. After loading the images to the custom designed Python APP, the recruitment intensity of each cell in the acquired image was generated and displayed in the graphs.

Materials and Methods:: Our Python APP replaced the manual processes previously done in ImageJ, streamlining the workflow and saving time on image processing and data analysis. Here's an overview of the steps in python APP:

1. 
   Load a series of images to the APP. The recruitment line showed up 2 min post of the laser cut as shown in Figure 1
   
   


2. 
   Use Cellpose, a deep learning-based image segmentation library, to perform cell detection and segmentation on a set of images as shown in Figure 2
   
   


3. 
   Crop the detected cells using the masks from Cellpose and save the cropped cell images in an output folder.
   
   


4. 
   Detect the brightest pixel in the cropped cell image using np.argmax() as the starting point for detecting the laser line.
   
   


5. 
   Create a binary mask by applying a threshold to the normalized image using the OpenCV library.
   
   


6. 
   Detect edges and contours in the binary mask and find the contour containing the brightest pixel, which represents the laser cut line in the cell, using the OpenCV library.
   
   


7. 
   Find the two farthest points in the contour and draw a line between them, representing the laser cut line as shown in Figure 3
   
   


8. 
   Calculate the brightness mean value of the laser cut line and write it into a CSV file.
   
   


9. 
   Plot the images and charts using the matplotlib library as shown in Figure 4
   
   

This program significantly improved the efficiency of the image processing and analysis, providing more accurate and faster results for studying the protein recruitment after laser damage




Results, Conclusions, and Discussions::

1. 
   We used Cellpose, a pretrained model, which can detect cells and perform cell segmentation with nearly 100% accuracy in our cell images. We will use Cellpose v2.0 to fine-tune our own data for further application.
   
   


2. 
   The Python program, incorporating machine-learning AI, replaced the manual image analysis previously done using ImageJ. This approach increased the accuracy by 60% and reduced the time spent on image processing and data collection by 83%.
   
   


3. 
   The cell images may contain unexpected bright spots that could interfere with the detection of the laser cut line. To address this issue, we developed an algorithm to detect and exclude those unexpected bright spots, ensuring the correct detection of the laser cut line. This innovation further improves the reliability and efficiency of our image processing and analysis methods.
   
   

Acknowledgements (Optional): :

References (Optional): : MouseLand. “Mouseland/Cellpose: A Generalist Algorithm for Cellular Segmentation with Human-in-the-Loop Capabilities.” GitHub, https://github.com/MouseLand/cellpose.