Merge pull request #11 from KlugerLab/10-check-gene_expression-matrix-is-positive

10 check gene expression matrix is positive

Author: fra-pcmgf

gene_trajectory/coarse_grain.py

@@ -4,6 +4,8 @@
 import scanpy as sc
 from sklearn.cluster import KMeans
 
+from gene_trajectory.util.input_validation import validate_matrix
+
 
 def select_top_genes(
         adata: sc.AnnData,
@@ -64,6 +66,10 @@ def coarse_grain(
     :param random_seed: the random seed
     :return: the updated cell embedding and gene expression matrices
     """
+    validate_matrix(gene_expression, obj_name='Gene Expression Matrix', min_value=0)
+    ncells, ngenes = gene_expression.shape
+    validate_matrix(cell_embedding, obj_name='Cell embedding', nrows=ncells)
+
     if cluster is None:
         k_means = KMeans(n_clusters=n, random_state=random_seed).fit(cell_embedding)
         cluster = k_means.labels_ # noqa

gene_trajectory/diffusion_map.py

@@ -1,6 +1,8 @@
 from typing import Union
 import numpy as np
 
+from gene_trajectory.util.input_validation import validate_matrix
+
 
 def diffusion_map(
         dist_mat: np.array,
@@ -19,6 +21,8 @@ def diffusion_map(
     :param t: Number of diffusion times
     :return: the diffusion embedding and the eigenvalues
     """
+    validate_matrix(dist_mat, square=True)
+
     affinity_matrix_symm = get_symmetrized_affinity_matrix(dist_mat=dist_mat, k=k, sigma=sigma)
     normalized_vec = np.sqrt(1 / affinity_matrix_symm.sum(axis=1))
     affinity_matrix_norm = (affinity_matrix_symm * normalized_vec * normalized_vec[:, None])
@@ -50,7 +54,8 @@ def get_symmetrized_affinity_matrix(
 
     :return:
     """
-    assert dist_mat.shape[0] == dist_mat.shape[1]
+    validate_matrix(dist_mat, square=True)
+
     dists = np.nan_to_num(dist_mat, 1e-6) # noqa
     k = min(k, dist_mat.shape[0])
 

gene_trajectory/extract_gene_trajectory.py

@@ -7,6 +7,7 @@
 import pandas as pd
 
 from gene_trajectory.diffusion_map import diffusion_map, get_symmetrized_affinity_matrix
+from gene_trajectory.util.input_validation import validate_matrix
 
 logger = logging.getLogger()
 
@@ -28,6 +29,8 @@ def get_gene_embedding(
     :param t: Number of diffusion times
     :return: the diffusion embedding and the eigenvalues
     """
+    validate_matrix(dist_mat, square=True)
+
     k = min(k, dist_mat.shape[0])
     n_ev = min(n_ev + 1, dist_mat.shape[0])
     diffu_emb, eigen_vals = diffusion_map(dist_mat=dist_mat, k=k, sigma=sigma, n_ev=n_ev, t=t)
@@ -47,6 +50,8 @@ def get_randow_walk_matrix(
     :param k: Adaptive kernel bandwidth for each point set to be the distance to its `K`-th nearest neighbor
     :return: Random-walk matrix
     """
+    validate_matrix(dist_mat, square=True)
+
     affinity_matrix_symm = get_symmetrized_affinity_matrix(dist_mat=dist_mat, k=k)
     normalized_vec = 1 / affinity_matrix_symm.sum(axis=1)
     affinity_matrix_norm = (affinity_matrix_symm * normalized_vec[:, None])
@@ -67,7 +72,7 @@ def get_gene_pseudoorder(
     :param max_id: Index of the terminal gene
     :return: The pseudoorder
     """
-    assert dist_mat.shape[0] == dist_mat.shape[1]
+    validate_matrix(dist_mat, square=True)
 
     emd = dist_mat[subset][:, subset]
     dm_emb, _ = diffusion_map(emd)
@@ -108,6 +113,8 @@ def extract_gene_trajectory(
     :param other: Label for genes not in a trajectory. Default: 'Other'
     :return: A data frame indicating gene trajectories and gene ordering along each trajectory
     """
+    validate_matrix(dist_mat, square=True)
+
     if np.isscalar(t_list):
         if n is None:
             raise ValueError(f'n should be specified if t_list is a number: {t_list}')

gene_trajectory/gene_distance_shared.py

@@ -9,6 +9,7 @@
 import ot
 from tqdm import tqdm
 
+from gene_trajectory.util.input_validation import validate_matrix
 from gene_trajectory.util.shared_array import SharedArray, PartialStarApply
 
 logger = logging.getLogger()
@@ -42,18 +43,20 @@ def cal_ot_mat(
     :return: the distance matrix
     """
     processes = int(processes) if isinstance(processes, float) else os.cpu_count()
-    n = gene_expr.shape[1]
+    validate_matrix(gene_expr, obj_name='Gene Expression Matrix', min_value=0)
+    ncells, ngenes = gene_expr.shape
+    validate_matrix(ot_cost, obj_name='Cost Matrix', shape=(ncells, ncells), min_value=0)
+
     if show_progress_bar:
         logger.info(f'Computing emd distance..')
 
     if gene_pairs is None:
-        pairs = ((i, j) for i in range(0, n - 1) for j in range(i + 1, n))
-        npairs = (n * (n - 1)) // 2
+        pairs = ((i, j) for i in range(0, ngenes - 1) for j in range(i + 1, ngenes))
+        npairs = (ngenes * (ngenes - 1)) // 2
     else:
         pairs = gene_pairs
         npairs = len(gene_pairs)
-
-    emd_mat = np.full((n, n), fill_value=np.NaN)
+    emd_mat = np.full((ngenes, ngenes), fill_value=np.NaN)
 
     with SharedMemoryManager() as manager:
         start_time = time.perf_counter()

gene_trajectory/util/input_validation.py

@@ -0,0 +1,70 @@
+from typing import Optional
+
+import numpy as np
+
+
+def validate_not_none(obj, obj_name: str = 'input'):
+    if obj is None:
+        raise ValueError(f"{obj_name} is None")
+
+
+def validate_matrix(
+        m: np.array,
+        obj_name: str = 'input',
+        nrows: Optional[int] = None,
+        ncols: Optional[int] = None,
+        shape: Optional[tuple[int, int]] = None,
+        square: Optional[bool] = None,
+        min_size: Optional[int] = 1,
+        min_value: Optional = None,
+        max_value: Optional = None,
+):
+    """
+    Validates an input matrix
+    @param m: the input matrix
+    @param obj_name: the name of the object, for error reporting
+    @param min_size: Minimum matrix size in each dimension. Defaults to 1, and will raise for an empty matrix
+    @param nrows: Number of rows in the matrix
+    @param ncols: Number of rows in the matrix
+    @param shape: the expected shape of the matrix
+    @param square: If True, ensures the matrix is square. If False, ensures the matrix is not
+    @param min_value: Minimum value for each element
+    @param max_value: Maximum value for each element
+    """
+    validate_not_none(m, obj_name=obj_name)
+
+    if len(m.shape) != 2:
+        raise ValueError(f"{obj_name} is not a matrix. Shape: {m.shape}")
+    mr, mc = m.shape
+
+    if nrows is not None:
+        if mr != nrows:
+            raise ValueError(f"{obj_name} does not have {nrows} rows. Shape: {m.shape}")
+
+    if ncols is not None:
+        if mc != ncols:
+            raise ValueError(f"{obj_name} does not have {ncols} columns. Shape: {m.shape}")
+
+    if shape is not None:
+        if m.shape != shape:
+            raise ValueError(f"{obj_name} does not have shape {shape}. Shape: {m.shape}")
+
+    if square is True:
+        if mr != mc:
+            raise ValueError(f"{obj_name} is not a square matrix. Shape: {m.shape}")
+    elif square is False:
+        if mr == mc:
+            raise ValueError(f"{obj_name} is a square matrix. Shape: {m.shape}")
+
+    if min_size is not None:
+        for s in m.shape:
+            if s < min_size:
+                raise ValueError(f"{obj_name} does not have enough elements. Min_size: {min_size}, Shape: {m.shape}")
+
+    if min_value is not None:
+        if m.min() < min_value:
+            raise ValueError(f"{obj_name} should not have values less than {min_value}. Minimum found: {m.min()}")
+
+    if max_value is not None:
+        if m.max() > max_value:
+            raise ValueError(f"{obj_name} should not have values greater than {max_value}. Maximum found: {m.max()}")

tests/test_compute_gene_distance_cmd.py

@@ -7,7 +7,7 @@
 from gene_trajectory.compute_gene_distance_cmd import cal_ot_mat
 
 
-class DiffusionMapTestCase(unittest.TestCase):
+class ComputeGeneDistanceTestCase(unittest.TestCase):
     gdm = np.array([
         [0, 1, 2],
         [1, 0, 2],

tests/test_gene_distance_shared.py

@@ -28,6 +28,16 @@ def test_gene_distance_shared(self):
         mt = cal_ot_mat(ot_cost=self.gdm, gene_expr=self.gem.T, show_progress_bar=False)
         np.testing.assert_almost_equal(self.expected_emd, mt, 6)
 
+    def test_gene_distance_input_validation(self):
+        with self.assertRaisesRegexp(ValueError, 'Cost Matrix does not have shape.*'):
+            cal_ot_mat(ot_cost=self.gdm, gene_expr=np.ones(shape=(6, 3)), show_progress_bar=False)
+
+        with self.assertRaisesRegexp(ValueError, 'Cost Matrix does not have shape.*'):
+            cal_ot_mat(ot_cost=np.ones(shape=(6, 3)), gene_expr=self.gem.T, show_progress_bar=False)
+
+        with self.assertRaisesRegexp(ValueError, 'Gene Expression Matrix should not have values less than 0.*'):
+            cal_ot_mat(ot_cost=np.ones(shape=(6, 3)), gene_expr=self.gem.T - 1, show_progress_bar=False)
+
     def test_cal_ot_mat_gene_pairs(self):
         exp = self.expected_emd.copy()
         exp[0, 2] = exp[2, 0] = 900

tests/test_input_validation.py

@@ -0,0 +1,29 @@
+import unittest
+
+import numpy as np
+
+from gene_trajectory.util.input_validation import validate_matrix
+
+
+class InputValidationTestCase(unittest.TestCase):
+    def test_validate_matrix(self):
+        m = np.array([[1, 2], [3, 4]])
+
+        validate_matrix(m, min_value=1, max_value=4, square=True, shape=(2, 2))
+
+        with self.assertRaisesRegexp(ValueError, '.*does not have 3 rows.*'):
+            validate_matrix(m, nrows=3)
+        with self.assertRaisesRegexp(ValueError, '.*does not have 8 columns.*'):
+            validate_matrix(m, ncols=8)
+        with self.assertRaisesRegexp(ValueError, '.*does not have shape \\(1, 1\\)'):
+            validate_matrix(m, shape=(1, 1))
+        with self.assertRaisesRegexp(ValueError, '.*Min_size: 3.*'):
+            validate_matrix(m, min_size=3)
+        with self.assertRaisesRegexp(ValueError, '.*should not have values less than 5.*'):
+            validate_matrix(m, min_value=5)
+        with self.assertRaisesRegexp(ValueError, '.*should not have values greater than 1.*'):
+            validate_matrix(m, max_value=1)
+
+
+if __name__ == '__main__':
+    unittest.main()