cellshape 0.0.15rc0

3D shape analysis using deep learning

3D single-cell shape analysis of cancer cells using geometric deep learning

This is a package for automatically learning and clustering cell shapes from 3D images. Please refer to our preprint on bioRxiv here

cellshape is available for everyone.

cellshape is the main package which is made up of sub-packages:

• cellshape-helper: https://github.com/Sentinal4D/cellshape-helper
• cellshape-cloud: https://github.com/Sentinal4D/cellshape-cloud
• cellshape-voxel: https://github.com/Sentinal4D/cellshape-voxel
• cellshape-cluster: https://github.com/Sentinal4D/cellshape-cluster

Installation and requirements

Dependencies

The software requires Python 3.7 or greater, PyTorch, torchvision, pyntcloud, numpy, scikit-learn, tensorboard, tqdm, datetime. This repo makes extensive use of cellshape-cloud, cellshape-cluster, cellshape-helper, and cellshape-voxel. to reproduce our results in our paper, only cellshape-cloud, cellshape-cluster are needed.

To install

1. We recommend creating a new conda environment:

conda create --name cellshape-env python=3.8
conda activate cellshape-env
pip install --upgrade pip

2. Install cellshape from pip

pip install cellshape

Hardware requirements

We have tested this software of an Ubuntu 20.04LTS with 128Gb RAM and NVIDIA Quadro RTX 6000 GPU.

Data structure

Our data is structured in the following way:

Data/
    all_data_stats.csv
    Plate1/
        stacked_pointcloud/
            Binimetinib/
                0010_0001_accelerator_20210315_bakal01_erk_main_21-03-15_12-37-27.ply
                ...
            Blebbistatin/
            ...
    Plate2/
        stacked_pointcloud/
    Plate3/
        stacked_pointcloud/

Data availability

Datasets to reproduce our results in our paper are available here.

Usage

The following steps assume that one already has point cloud representations of cells or nuclei. If you need to generate point clouds from 3D binary masks please go to cellshape-helper.

The training procedure follows two steps:

1. Training the dynamic graph convolutional foldingnet (DFN) autoencoder to automatically learn shape features.
2. Adding the clustering layer to refine shape features and learn shape classes simultaneous.

Inference can be done after each step.

1. Train DFN autoencoder

python 

import torch
from torch.utils.data import DataLoader
from datetime import datetime
import logging

import cellshape_cloud as cscloud
import cellshape_cluster as cscluster
from cellshape_cloud.vendor.chamfer_distance import ChamferLoss
from cellshape_cloud.helpers.reports import get_experiment_name


input_dir = "/home/mvries/Documents/CellShape/DatasetForTesting/"
batch_size = 20
learning_rate_autoencoder = 0.00001
learning_rate_clustering = 0.000001
num_features = 128
num_clusters = 3
num_epochs_autoencoder = 1
num_epochs_clustering = 3
k=20
encoder_type="dgcnn"
decoder_type = "foldingnetbasic"
output_dir = "/home/mvries/Documents/Testing_output/"
gamma = 1
alpha = 1.0
divergence_tolerance = 0.01
update_interval = 1


autoencoder = cscloud.CloudAutoEncoder(num_features=num_features, 
                         k=k,
                         encoder_type=encoder_type,
                         decoder_type=decoder_type)

dataset = cscloud.PointCloudDataset(input_dir)

dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=True)

criterion = ChamferLoss()

optimizer = torch.optim.Adam(
    autoencoder.parameters(),
    lr=learning_rate_autoencoder * 16 / batch_size,
    betas=(0.9, 0.999),
    weight_decay=1e-6,
)

name_logging, name_model, name_writer, name = get_experiment_name(
        model=autoencoder, output_dir=output_dir
    )

logging_info = name_logging, name_model, name_writer, name

now = datetime.now().strftime("%d/%m/%Y %H:%M:%S")
logging.basicConfig(filename=name_logging, level=logging.INFO)
logging.info(f"Started training model {name} at {now}.")

output_cloud = cscloud.train(autoencoder, 
                             dataloader,
                             num_epochs_autoencoder, 
                             criterion, 
                             optimizer,
                             logging_info)

autoencoder = output_cloud[0]


model = cscluster.DeepEmbeddedClustering(autoencoder=autoencoder, 
                               num_clusters=num_clusters)

dataloader = DataLoader(dataset, batch_size=batch_size, shuffle=False) 
# it is very important that shuffle=False here!
dataloader_inf = DataLoader(dataset, batch_size=1, shuffle=False) 
# it is very important that batch_size=1 and shuffle=False here!

optimizer = torch.optim.Adam(
    model.parameters(),
    lr=learning_rate_clustering * 16 / batch_size,
    betas=(0.9, 0.999),
    weight_decay=1e-6,
)

reconstruction_criterion = ChamferLoss()
cluster_criterion = torch.nn.KLDivLoss(reduction="sum")

cscluster.train(
    model,
    dataloader,
    dataloader_inf,
    num_epochs_clustering,
    optimizer,
    reconstruction_criterion,
    cluster_criterion,
    update_interval,
    gamma,
    divergence_tolerance,
    logging_info
)

For developers

• Fork the repository
• Clone your fork

git clone https://github.com/USERNAME/cellshape

• Install an editable version (-e) with the development requirements (dev)

cd cellshape
pip install -e .[dev] 

• To install pre-commit hooks to ensure formatting is correct:

pre-commit install

• To release a new version:

Firstly, update the version with bump2version (bump2version patch, bump2version minor or bump2version major). This will increment the package version (to a release candidate - e.g. 0.0.1rc0) and tag the commit. Push this tag to GitHub to run the deployment workflow:

git push --follow-tags

Once the release candidate has been tested, the release version can be created with:

bump2version release

References

[1] An Tao, 'Unsupervised Point Cloud Reconstruction for Classific Feature Learning', GitHub Repo, 2020