Merge pull request #30 from VasudhaJha/feature/export-all-files

Feature/export all files

.github/workflows/release.yml

@@ -14,5 +14,5 @@ jobs:
     steps:
       - uses: rymndhng/release-on-push-action@master
         with:
-          bump_version_scheme: patch # can be either "major", "minor", "patch" or "norelease"
+          bump_version_scheme: minor # can be either "major", "minor", "patch" or "norelease"
           tag_prefix: v

README.md

@@ -41,6 +41,7 @@ plt.figure(1) # Plot the first genomap
 plt.imshow(findI, origin = 'lower', extent = [0, 10, 0, 10], aspect = 1)
 plt.title('Genomap of a cell from TM dataset')
 ```
+
 ### Example 2 - Try genoVis for data visualization and clustering
 
 ```python
@@ -274,4 +275,3 @@ Islam, M.T., Xing, L. Cartography of Genomic Interactions Enables Deep Analysis
 
 
 
-

genomap/genoClassification/__init__.py

@@ -0,0 +1 @@
+from .genoClassification import *

genomap/genoClassification/genoClassification.py

@@ -0,0 +1,65 @@
+
+"""
+Created on Sun Jul 16 21:27:17 2023
+@author: Md Tauhidul Islam, Research Scientist, Dept. of radiation Oncology, Stanford University
+"""
+
+import numpy as np
+import os
+import torch
+from torch.utils.data import DataLoader, Dataset
+import scipy
+
+from genomap.genomap import construct_genomap
+from genomap.utils.util_Sig import select_n_features
+import genomap.genoNet as gNet
+
+def genoClassification(training_data, training_labels, test_data, rowNum=32, colNum=32, epoch=100):
+
+    # training_data: numpy array in cellxgene shape
+    # training_labels: numpy array 
+    # test_data: numpy array in cellxgene shape
+    # rowNum: row number of genomap
+    # colNum: column number of genomap
+    # epoch: epoch number for training
+
+    data = np.concatenate((training_data, test_data), axis=0)
+    # Construction of genomaps
+    nump=rowNum*colNum 
+    if nump<data.shape[1]:
+        data,index=select_n_features(data,nump)
+
+    dataNorm=scipy.stats.zscore(data,axis=0,ddof=1)
+    genoMaps=construct_genomap(dataNorm,rowNum,colNum,epsilon=0.0,num_iter=200)    
+
+    # Split the data for training and testing
+    dataMat_CNNtrain = genoMaps[:training_labels.shape[0]] 
+    dataMat_CNNtest = genoMaps[training_labels.shape[0]:]
+
+    groundTruthTrain = training_labels
+    classNum = len(np.unique(groundTruthTrain))
+
+    # Preparation of training and testing data for PyTorch computation
+    XTrain = dataMat_CNNtrain.transpose([0,3,1,2])
+    XTest = dataMat_CNNtest.transpose([0,3,1,2])
+    yTrain = groundTruthTrain
+
+    # Train the network in PyTorch framework
+    miniBatchSize = 128
+    net = gNet.traingenoNet(XTrain, yTrain, maxEPOCH=epoch, batchSize=miniBatchSize, verbose=True)
+
+    # # Process the test data
+    yTest = np.random.rand(dataMat_CNNtest.shape[0])
+    testset = gNet.geneDataset(XTest, yTest)
+    testloader = gNet.DataLoader(testset, batch_size=miniBatchSize, shuffle=False)
+    device = gNet.get_device()
+
+    # Perform data classification
+    prob_test = gNet.predict(net, testloader, device)
+    pred_test = np.argmax(prob_test, axis=-1)
+
+    return pred_test
+
+
+
+

genomap/genoDR/__init__.py

@@ -0,0 +1 @@
+from .genoDimReduction import *

genomap/genoDR/genoDimReduction.py

@@ -0,0 +1,67 @@
+# -*- coding: utf-8 -*-
+"""
+Created on Wed Jul 12 16:11:15 2023
+
+@author: Md Tauhidul Islam, Ph.D., Dept. of Radiation Oncology, Stanford University
+"""
+import numpy as np
+import pandas as pd
+import scanpy as sc
+from sklearn.feature_selection import VarianceThreshold
+from tensorflow.keras.optimizers import Adam
+import umap
+
+from genomap.genomap import construct_genomap
+from genomap.utils.ConvIDEC import ConvIDEC
+from genomap.utils.gTraj_utils import nearest_divisible_by_four
+from genomap.utils.util_Sig import select_n_features
+
+def genoDR(data,n_dim=32, n_clusters=None, colNum=32,rowNum=32,batch_size=64,verbose=1,
+                    pretrain_epochs=100,maxiter=300):
+
+# Construction of genomap    
+    colNum=nearest_divisible_by_four(colNum)
+    rowNum=nearest_divisible_by_four(rowNum)
+    nump=rowNum*colNum 
+    if nump<data.shape[1]:
+        data,index=select_n_features(data,nump)
+
+    genoMaps=construct_genomap(data,rowNum,colNum,epsilon=0.0,num_iter=200)
+
+    if n_clusters==None:
+
+        adata = convertToAnnData(data)
+        # Perform Louvain clustering
+        sc.pp.neighbors(adata)
+        sc.tl.louvain(adata, resolution=1.0)
+        # Access the cluster assignments
+        cluster_labels = adata.obs['louvain']
+        n_clusters = len(np.unique(cluster_labels)) 
+
+# Deep learning-based dimensionality reduction and clustering
+    optimizer = Adam()    
+    model = ConvIDEC(input_shape=genoMaps.shape[1:], filters=[32, 64, 128, n_dim], n_clusters=n_clusters)
+    model.compile(optimizer=optimizer, loss=['kld', 'mse'], loss_weights=[0.1, 1.0])
+    pretrain_optimizer ='adam'
+    update_interval=50
+    model.pretrain(genoMaps, y=None, optimizer=pretrain_optimizer, epochs=pretrain_epochs, batch_size=batch_size,
+                        verbose=verbose)
+
+    y_pred = model.fit(genoMaps, y=None, maxiter=maxiter, batch_size=batch_size, update_interval=update_interval,
+                       )
+    y_pred = model.predict_labels(genoMaps)
+    feat_DNN=model.extract_features(genoMaps)
+    #embedding2D = umap.UMAP(n_neighbors=30,min_dist=0.3,n_epochs=200).fit_transform(feat_DNN)
+    return feat_DNN
+
+
+def convertToAnnData(data):
+    # Create pseudo cell names
+    cell_names = ['Cell_' + str(i) for i in range(1, data.shape[0]+1)]
+    # Create pseudo gene names
+    gene_names = ['Gene_' + str(i) for i in range(1, data.shape[1]+1)]
+    # Create a pandas DataFrame
+    df = pd.DataFrame(data, index=cell_names, columns=gene_names)
+    adataMy=sc.AnnData(df)
+    return adataMy
+

genomap/genoMOI/genoMOI.py

@@ -6,14 +6,15 @@
 """
 
 from tensorflow.keras.optimizers import Adam
-from genomap.utils import ConvIDEC
+from genomap.utils.ConvIDEC import ConvIDEC
 from sklearn.feature_selection import VarianceThreshold
 from genomap.genomap import construct_genomap
 import umap
 from genomap.utils.gTraj_utils import nearest_divisible_by_four
-from genomap.utils.utils_MOI import *
+from genomap.utils.utils_MOI import * 
+from genomap.utils.util_Sig import select_n_features
 
-def compute_genoMOI(*arrays,n_clusters=None, colNum, rowNum):  
+def genoMOI(*arrays,n_clusters=None, colNum, rowNum):  
 
 # arrays: number of arrays such as array1,array2
 # n_clusters: number of data classes
@@ -50,12 +51,7 @@ def extract_genoVis_features(data,n_clusters=20, colNum=32,rowNum=32,batch_size=
     rowNum=nearest_divisible_by_four(rowNum)
     nump=rowNum*colNum 
     if nump<data.shape[1]:
-    # create an instance of the VarianceThreshold class
-        selector = VarianceThreshold( )
-    # fit the selector to the data and get the indices of the top n most variable features
-        var_threshold = selector.fit(data)
-        top_n_indices = var_threshold.get_support(indices=True)
-        data=data[:,top_n_indices[0:nump]]  
+        data,index=select_n_features(data,nump) 
     genoMaps=construct_genomap(data,rowNum,colNum,epsilon=0.0,num_iter=200)
 
 # Deep learning-based dimensionality reduction and clustering

genomap/genoNet/__init__.py

@@ -0,0 +1 @@
+from .genoNet import *

genomap/genoNet/genoNet.py

@@ -1,9 +1,7 @@
-# -*- coding: utf-8 -*-
-"""
-Supervised genoNet
-Created on Sun May 15 09:03:29 2022
 
-@author: anonymous
+"""
+Created on Sun Jul 16 21:27:17 2023
+@author: Md Tauhidul Islam, Research Scientist, Dept. of radiation Oncology, Stanford University
 """
 
 import os

genomap/genoNetRegression/__init__.py

@@ -0,0 +1 @@
+from .genoNet_regression import *