AI_SEC

This study utilizes different machine learning and deep learning models to provide a proof of concept that organismic information at a macroscopic length scale may be de-encrypted from a snapshot of only a few dozen cells.

• Paper:
• Data:

Notifications

• Do NOT modify the name of this repository. If you really need to rename the repository, remember to update the list of strings in utils.py.
• At least 150 GB of disk space is required.
• Do NOT run a script with a larger number before ensuring that all scripts with smaller numbers have been executed, i.e., always execute scripts in sequential order.
• Create the dataset on an SSD or using an OS like Linux with aggressive caching to ensure reasonable processing times.
• The entire project uses glob for file searches. If you need to save images in the project folder, please use file formats other than .tif or .tiff.

System Information

• CPU: i9-13900KS
• GPU: RTX-4090 with 24GB VRAM
• RAM: 64 GB DDR5
• OS: Ubuntu 22.04
• Python Version: 3.10

Data

The data is available from DOI:10.6019/S-BIAD1654

Installation

1. Follow the instructions here to install Miniforge.

2. Create a new environment and activate it:

   mamba create -n AI_SEC python=3.10
   conda activate AI_SEC

3. Install the required packages (Tested on 2024-06-03):

   mamba install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia
   pip install scikit-image==0.22.0 scikit-learn==1.4.0
   pip install -U colorama toml tomlkit matplotlib tqdm rich seaborn imagecodecs
   mamba install pandas pyimagej openjdk=8 imgaug=0.4.0
   pip install albumentations==1.3.1
   pip install grad-cam==1.4.8
   pip install umap-learn==0.5.6
   mamba install numpy=1.23.0
   mamba install mkl==2024.0

4. Download the Fiji (ImageJ) and udpate the absolute path to db_path_plan.toml (see the db_path_plan.toml section below).

5. Install the Fiji (ImageJ) plugin Find Focused Slices

   1. Download the plugin file Find_focused_slices.class

   2. Place it into the plugins folder of your Fiji (ImageJ) directory:

      📂 Fiji.app/
      ├── 📂 ...
      ├── 📂 plugins/
      │   ├── 📂 ...
      │   ├── 📄 Find_focused_slices.class (◀️ Place the file here)
      │   └── 📄 ...
      └── ...
      

6. Download and unzip our data, then organize it into the following structure:

   📂 AISEC.data/
   ├── 📂 {Data}_Processed/
   ├── 📂 {Dataset}_DL/
   ├── 📂 {Dataset}_ML/
   ├── 📂 {Model}_BFSeg/
   ├── 📂 {Model}_Cellpose/
   ├── 📂 {Model}_DL/
   ├── 📂 {Results}_Advanced/ (◀️ manually create the empty folder)
   ├── 📂 {Results}_DL/ (◀️ manually create the empty folder)
   └── 📂 {Results}_ML/ (◀️ manually create the empty folder)
   

   Note: To avoid encountering a FileNotFoundError, manually create the {Results}_Advanced/, {Results}_DL/, and {Results}_ML/ directories after unzipping.

7. Update the absolute path in db_path_plan.toml (see the db_path_plan.toml section below).

File Structure

Tools/ : The scripts are for debugging. Please do not run them unless necessary; use at your own risk.

dev/ : The files are under development or already deprecated, so they may not work. Please do not run them unless necessary; use at your own risk.

modules/ : All the utility functions and classes.

script_data/ : Converts .lif files into commonly used image formats, splits into train/validation/test sets, applies K-Means, etc.

script_bfseg/ : ResNet18-UNet automated segmentation for obtaining the "trunk surface area". Inspired by usuyama/pytorch-unet.

script_ml/ : From Cellpose segmentation to simple features, machine learning results, and UMAP. Inspired by ccsyan/labeling-cells-using-slic.

script_dl/ : Everything related to deep learning, including CAM analysis.

script_adv/ : Advanced results, such as "critical SEC number" and "feature-subtracted images", etc.

Usage

** For detailed instructions on each part, please click on the corresponding directory (folder) listed in the section above.

db_path_plan.toml

The TOML file manages the linkage between data and code, allowing them to be stored separately. This setup enables you to place the our data anywhere on your system.

It is parsed as a directory within the program:

• The left side (key) represents the variable used in the code.
• The right side (value) represents the name of the corresponding data directory.

root = "" # absolute path of the data directory on this computer

data_processed      = "{Data}_Processed"
dataset_cropped_v3  = "{Dataset}_DL"
dataset_ml          = "{Dataset}_ML"
model_bfseg         = "{Model}_BFSeg"
model_cellpose      = "{Model}_Cellpose"
model_history       = "{Model}_DL"
model_prediction    = "{Results}_DL"
result_ml           = "{Results}_ML"
result_adv          = "{Results}_Advanced"

fiji_local = "" # absolute path of "Fiji (ImageJ)" on this computer

• The root key should be assigned to the absolute path of the data directory on your system.
• The fiji_local key should be assigned to the absolute path of Fiji (ImageJ) on your system.
• The names of the data directories are adjustable. If you change them, remember to update the corresponding names in the TOML file.

Note: Key-value pairs that appear in the TOML file but are not mentioned here do not affect your experience with this project.

Thanks

• https://sites.google.com/site/qingzongtseng/find-focus
• https://github.com/ccsyan/labeling-cells-using-slic
• https://github.com/usuyama/pytorch-unet