Tutorial3: scDREAMER-Sup semi-supervised integration of Healthy Heart dataset

Here we present tutorial on semi-supervised integration of Healthy Heart dataset. The source of the dataset is https://www.heartcellatlas.org/ The heart cells are obtained in 147 batches corresponding to different donors, sequential protocols, race, gender, age range, and death conditions. Total 12 major cardiac cell types are assigned in this 486134 cells dataset and ~34000 cells are not assigned any cell label.

The Tutorial demonstrates scDREAMER-Sup’s semi-supervised integration of large number of batches.

Importing Libraries

import warnings
warnings.filterwarnings('ignore')

import os
import scanpy as sc
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
import random
import numpy as np
import tensorflow as tf2
import tensorflow.compat.v1 as tf
tf.disable_v2_behavior()

2023-09-26 15:05:37.471577: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX512F FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2023-09-26 15:05:38.088576: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT

WARNING:tensorflow:From /home/ajita/anaconda3/lib/python3.9/site-packages/tensorflow/python/compat/v2_compat.py:107: disable_resource_variables (from tensorflow.python.ops.variable_scope) is deprecated and will be removed in a future version.
Instructions for updating:
non-resource variables are not supported in the long term

Visualization of Healthy Heart data

adata = sc.read_h5ad("/home/ajita/Documents/data_integration/Healthy_Heart/Healthy_human_heart_adata_subset.h5ad")
sc.pp.neighbors(adata, use_rep = "X")
sc.tl.umap(adata)
sc.pl.umap(adata, color = "celltype", frameon = False)

# Diversity of Samples
sc.pl.umap(adata, color = ["TissueDetail", "protocol"], frameon = False)
sc.pl.umap(adata, color = ["Race", "DeathType"], frameon = False)
sc.pl.umap(adata, color = ["Gender", "SampleType"], frameon = False)
sc.pl.umap(adata, color = ["DonorID", "AgeBin"], frameon = False)

Setting seed for reproducibility

np.random.seed(0)
tf.set_random_seed(0)
random.seed(0)
tf2.random.set_seed(0)
tf2.keras.utils.set_random_seed(0)

Building model

name = "Healthy_Heart"

"""
NOTE:
Run setting as follows:
0: Supervised setting: In case of healthy heart, semi-supervised setting for "Not-assigned" cells
10: 10 percent missing labels data
20: 20 percent missing labels data
50: 50 percent missing labels data
"""
run_setting = 0

Run setting

path = "/home/ajita/Documents/data_integration/"

data_path = {

"Healthy_Heart" : {0: path + "Healthy_Heart/Healthy_human_heart_adata_subset.h5ad"
                 }
}


batch_key_dict = {'Immune_Human' : 'batch',
                 "Healthy_Heart" : "sampleID"

                }

"""
In semi-supervised setting,  cells with cell type labels "NA" will not
participate in training the cell-type classifier"

"""

cell_type_key_dict = {
    "Healthy_Heart" : {0: "celltype_NA" }
}

lr_big_data = {"lr_ae" : 0.0001, "lr_dis": 0.00001}

learning_rate = {
"Healthy_Heart" : {0: lr_big_data, 20: lr_big_data, 50: lr_big_data}}


plot_cell_type_dict = {
                    "Healthy_Heart" : "celltype",

                    }

scDREAMER model

from scDREAMER import scDREAMER_SUP


run_config = tf.ConfigProto()

run_config.gpu_options.per_process_gpu_memory_fraction = 0.333
run_config.gpu_options.allow_growth = True

with tf.Session(config = run_config) as sess:

    dreamer = scDREAMER_SUP(
        sess,
        epoch = 300,
        dataset_name = data_path[name][run_setting],
        batch = batch_key_dict[name],
        cell_type = cell_type_key_dict[name][run_setting],
        plot_cell_type = plot_cell_type_dict[name],
        name = name,
        lr_ae = learning_rate[name][run_setting]['lr_ae'],
        lr_dis = learning_rate[name][run_setting]['lr_dis']
        )

    dreamer.train_cluster()

2023-09-21 20:04:59.946007: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1635] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 5367 MB memory:  -> device: 0, name: Quadro RTX 5000, pci bus id: 0000:d8:00.0, compute capability: 7.5

Loading dataset
Preprocessing...
here [10 11 11 ...  2  2  2]
Shape self.data_train: (485511, 2000)
Shape self.data_test: (485511, 2000)
encoder input shape  Tensor("concat:0", shape=(?, 2147), dtype=float32)
WARNING:tensorflow:From /home/ajita/anaconda3/lib/python3.9/site-packages/tensorflow/python/util/dispatch.py:1176: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.
Instructions for updating:
Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.
WARNING:tensorflow:From /home/ajita/anaconda3/lib/python3.9/site-packages/tensorflow/python/util/dispatch.py:1176: to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.cast` instead.
decoder input shape  Tensor("concat_2:0", shape=(?, 157), dtype=float32)
KL gaussian z Tensor("mul_12:0", shape=(?,), dtype=float32)
KL gaussian l Tensor("mul_11:0", shape=(?,), dtype=float32)
WARNING:tensorflow:From /home/ajita/anaconda3/lib/python3.9/site-packages/tensorflow/python/util/dispatch.py:1176: softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version.
Instructions for updating:

Future major versions of TensorFlow will allow gradients to flow
into the labels input on backprop by default.

See `tf.nn.softmax_cross_entropy_with_logits_v2`.

scDREAMER-Sup on DataSet /home/ajita/Documents/data_integration/Healthy_Heart/Healthy_human_heart_adata_subset.h5ad ...

2023-09-21 20:06:58.537832: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:353] MLIR V1 optimization pass is not enabled

Epoch : [0] ,  a_loss = 217.2630
Epoch : [10] ,  a_loss = 194.5482
Epoch : [20] ,  a_loss = 189.7468
Epoch : [30] ,  a_loss = 187.8285
Epoch : [40] ,  a_loss = 186.5633
Epoch : [50] ,  a_loss = 186.2601
Epoch : [60] ,  a_loss = 185.5387
Epoch : [70] ,  a_loss = 184.7160
Epoch : [80] ,  a_loss = 184.6446
Epoch : [90] ,  a_loss = 184.2780
Epoch : [100] ,  a_loss = 183.8163
Epoch : [110] ,  a_loss = 183.4391
Epoch : [120] ,  a_loss = 182.9392
Epoch : [130] ,  a_loss = 182.6407
Epoch : [140] ,  a_loss = 181.9946
Epoch : [150] ,  a_loss = 181.7168
Epoch : [160] ,  a_loss = 181.2697
Epoch : [170] ,  a_loss = 180.9640
Epoch : [180] ,  a_loss = 180.4101
Epoch : [190] ,  a_loss = 180.0579
Epoch : [200] ,  a_loss = 179.8580
Epoch : [210] ,  a_loss = 179.6591
Epoch : [220] ,  a_loss = 179.5249
Epoch : [230] ,  a_loss = 179.5214
Epoch : [240] ,  a_loss = 179.6433
Epoch : [250] ,  a_loss = 179.9576
Epoch : [260] ,  a_loss = 179.8742
Epoch : [270] ,  a_loss = 180.0580
Epoch : [280] ,  a_loss = 180.1403
Epoch : [290] ,  a_loss = 180.2992
latent_matrix shape (485511, 10)
(485511,)

None

Done !