cellshape-cluster

---

Cellshape-cluster is an easy-to-use tool to analyse the cluster cells by their shape using deep learning and, in particular, deep-embedded-clustering. The tool provides the ability to train popular graph-based or convolutional autoencoders on point cloud or voxel data of 3D single cell masks as well as providing pre-trained networks for inference.

To install

pip install cellshape-cluster

Usage

Basic usage:

import torch
from cellshape_cloud import CloudAutoEncoder
from cellshape_cluster import DeepEmbeddedClustering

autoencoder = CloudAutoEncoder(
    num_features=128, 
    k=20, 
    encoder_type="dgcnn"
)

model = DeepEmbeddedClustering(autoencoder=autoencoder, 
                               num_clusters=10,
                               alpha=1.0)

points = torch.randn(1, 2048, 3)

recon, features, clusters = model(points)

To load a trained graph-based autoencoder and perform deep embedded clustering:

import torch
from torch.utils.data import DataLoader

import cellshape_cloud as cloud
import cellshape_cluster as cluster
from cellshape_cloud.vendor.chamfer_distance import ChamferDistance

dataset_dir = "path/to/pointcloud/dataset/"
autoencoder_model = "path/to/autoencoder/model.pt"
num_features = 128
k = 20
encoder_type = "dgcnn"
num_clusters = 10
num_epochs = 1
learning_rate = 0.00001
gamma = 1
divergence_tolerance = 0.01
output_dir = "path/to/output/"


autoencoder = CloudAutoEncoder(
    num_features=128, 
    k=20, 
    encoder_type="dgcnn"
)

checkpoint = torch.load(autoencoder_model)

autoencoder.load_state_dict(checkpoint['model_state_dict']

model = DeepEmbeddedClustering(autoencoder=autoencoder, 
                               num_clusters=10,
                               alpha=1.0)

dataset = cloud.PointCloudDataset(dataset_dir)

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 * 16 / batch_size,
    betas=(0.9, 0.999),
    weight_decay=1e-6,
)

reconstruction_criterion = ChamferDistance()
cluster_criterion = nn.KLDivLoss(reduction="sum")

train(
    model,
    dataloader,
    dataloader_inf,
    num_epochs,
    optimizer,
    reconstruction_criterion,
    cluster_criterion,
    update_interval,
    gamma,
    divergence_tolerance,
    output_dir
)

Parameters

• autoencoder: CloudAutoEncoder or VoxelAutoEncoder.
  Instance of autoencoder class from cellshape-cloud or cellshape-voxel
• num_clusters: int.
  The number of clusters to use in deep embedded clustering algorithm.
• alpha: float.
  Degrees of freedom for the Student's t-distribution. Xie et al. (ICML, 2016) let alpha=1 for all experiments.