make palantir trsults fully reproducible

src/palantir/core.py

@@ -6,7 +6,6 @@
 import pandas as pd
 import networkx as nx
 import time
-import random
 import copy
 
 from sklearn.metrics import pairwise_distances
@@ -45,6 +44,7 @@ def run_palantir(
     entropy_key: str = "palantir_entropy",
     fate_prob_key: str = "palantir_fate_probabilities",
     waypoints_key: str = "palantir_waypoints",
+    seed: int = 20,
 ) -> Optional[PResults]:
     """
     Executes the Palantir algorithm to derive pseudotemporal ordering of cells, their fate probabilities, and
@@ -81,6 +81,8 @@ def run_palantir(
         Column names of the probability matrix are stored in the AnnData's uns[fate_prob_key + "_columns"].
     waypoints_key : str, optional
         Key to store the waypoints in uns of the AnnData object. Default is 'palantir_waypoints'.
+    seed : int, optional
+        The seed for the random number generator used in waypoint sampling. Default is 20.
 
     Returns
     -------
@@ -126,7 +128,7 @@ def run_palantir(
 
     # Append start cell
     if isinstance(num_waypoints, int):
-        waypoints = _max_min_sampling(data_df, num_waypoints)
+        waypoints = _max_min_sampling(data_df, num_waypoints, seed)
     else:
         waypoints = num_waypoints
     waypoints = waypoints.union(dm_boundaries)
@@ -175,18 +177,20 @@ def run_palantir(
     return pr_res
 
 
-def _max_min_sampling(data, num_waypoints):
+def _max_min_sampling(data, num_waypoints, seed=None):
     """Function for max min sampling of waypoints
 
     :param data: Data matrix along which to sample the waypoints,
                  usually diffusion components
     :param num_waypoints: Number of waypoints to sample
-    :param num_jobs: Number of jobs for parallel processing
+    :param seed: Random number generator seed to find initial guess.
     :return: pandas Series reprenting the sampled waypoints
     """
 
     waypoint_set = list()
     no_iterations = int((num_waypoints) / data.shape[1])
+    if seed is not None:
+        np.random.seed(seed)
 
     # Sample along each component
     N = data.shape[0]
@@ -195,7 +199,9 @@ def _max_min_sampling(data, num_waypoints):
         vec = np.ravel(data[ind])
 
         # Random initialzlation
-        iter_set = random.sample(range(N), 1)
+        iter_set = [
+            np.random.randint(N),
+        ]
 
         # Distances along the component
         dists = np.zeros([N, no_iterations])
@@ -312,6 +318,7 @@ def identify_terminal_states(
     num_waypoints=1200,
     n_jobs=-1,
     max_iterations=25,
+    seed=20,
 ):
 
     # Scale components
@@ -330,7 +337,7 @@ def identify_terminal_states(
 
     # Sample waypoints
     # Append start cell
-    waypoints = _max_min_sampling(data, num_waypoints)
+    waypoints = _max_min_sampling(data, num_waypoints, seed)
     waypoints = waypoints.union(dm_boundaries)
     waypoints = pd.Index(waypoints.difference([start_cell]).unique())
 

src/palantir/presults.py

@@ -256,7 +256,7 @@ def compute_gene_trends(
     pseudo_time = ad.obs[pseudo_time_key].values
     masks = ad.obsm[masks_key]
     branches = ad.uns[masks_key + "_columns"]
-    
+
     if lineages is not None:
         for lin in lineages:
             if lin not in branches: