feat: add mb-PHENIX as a scikit estimator

CODE/mb-phenix code/mb_phenix_scaler.py

@@ -0,0 +1,108 @@
+import pandas as pd
+import numpy as np
+from typing import List
+from sklearn.base import BaseEstimator, TransformerMixin
+
+import umap
+from scipy.spatial.distance import pdist
+from scipy.spatial.distance import squareform
+
+class MBPhenixScaler(BaseEstimator, TransformerMixin):
+  def __init__(self, ignored_features: List[str] = [], target_weight = 0.9):
+    self.columns = None
+    self.ignored_features = ignored_features
+    self.scaler = umap.UMAP(n_components=10, verbose=False,metric='cosine',n_epochs=1000,min_dist=0.1,n_neighbors=200,
+                                random_state=123, target_weight=target_weight)
+    self.new_matrix = None
+
+  def fit(self, X, y=None):
+      """
+      Fits the scaler to the data.
+
+      Parameters:
+        X (pandas.DataFrame): The input data to fit the scaler on.
+        y (pandas.Series or None): The target variable. Default is None. Strongly recommend supervised learning.
+
+      Returns:
+        self (Scaler): The fitted scaler object.
+      """
+      relevant_columns = X.loc[:, ~X.columns.isin(self.ignored_features)]
+      self.columns = [c for c in relevant_columns.columns]
+      self._fit_not_ignored(relevant_columns, y)
+
+      return self
+
+  def _fit_not_ignored(self, X, y=None, t=3, decay=1, metric='euclidean', knn=10):
+    umap_data = self.scaler.fit_transform(X.values, y)
+
+    distance_matrix =pdist(umap_data, metric)
+    distance_matrix = (squareform(distance_matrix))
+    D = distance_matrix
+    n,m = D.shape
+    E = np.zeros((m,m))
+
+    knn_dst = np.sort(distance_matrix, axis=1)
+
+    epsilon = knn_dst[:,knn]
+    pdx_scale = (distance_matrix / epsilon).T
+
+    E = np.exp(-1 * ( pdx_scale ** decay))
+    A = (E + E.T)
+
+    diff_deg = np.diag(np.sum(A,0))
+    diff_op = np.dot(np.diag(np.diag(diff_deg)**(-1)),A)
+
+    self.new_matrix =  np.linalg.matrix_power(diff_op, t)
+
+  def transform(self, X) -> pd.DataFrame:
+    """Transforms the features.
+
+    Arguments:
+      X (pd.DataFrame): The data to be transformed.
+
+    Returns:
+      pd.DataFrame: The transformed data.
+    """
+    missing_columns = [c for c in self.columns if c not in list(X.columns)]
+
+    if len(missing_columns) / len(self.columns) > 0.2:
+        print(f'More than 20 percent of columns not found in  {self.__class__}. {len(X.columns)} available columns: {list(X.columns)}.')
+
+    X = X.copy()
+
+    for m in missing_columns:
+      X[m] = 0.0
+
+    relevant_columns = X.loc[:, ~X.columns.isin(self.ignored_features)].copy()
+    relevant_columns = relevant_columns[self.columns]
+
+    return self._transform_not_ignored(relevant_columns).join(X.loc[:, X.columns.isin(self.ignored_features)])
+
+  def _transform_not_ignored(self, X) -> pd.DataFrame:
+    """Uses mb-PHENIX to transfrom features. (DOI: 10.1093/bioinformatics/btad706)
+
+    Arguments:
+      X (pd.DataFrame): The data to be transformed.
+
+    Returns:
+      pd.DataFrame: The transformed data.
+    """
+
+    data_new = np.array(np.dot(self.new_matrix, X))
+
+    Matix_col_genes_row_cell2 = (X +1) - X
+    Matix_col_genes_row_cell2 = Matix_col_genes_row_cell2 - Matix_col_genes_row_cell2
+    Matix_impu = Matix_col_genes_row_cell2 + data_new
+
+    sc_PHENIX = Matix_impu 
+
+    return (sc_PHENIX)
+
+
+  def get_support(self) -> List[str]:
+    """Returns a list of selected features
+
+    Returns:
+        List[str]: list of selected features
+    """
+    return self.columns