diff --git a/scdna_replication_tools/pert_model.py b/scdna_replication_tools/pert_model.py
index 05d9bc7..e3264bf 100644
--- a/scdna_replication_tools/pert_model.py
+++ b/scdna_replication_tools/pert_model.py
@@ -187,8 +187,18 @@ def process_input_data(self):
             assert cn_s_reads.shape[0] == gammas.shape[0] == rt_prior_profile.shape[0]
         else:
             rt_prior_profile = None
+        
+        # get tensor for rt initialization if provided
+        if (self.rt_init_col is not None) and (self.rt_init_col in self.cn_s.columns):
+            rt_init_profile = self.cn_s[[self.chr_col, self.start_col, self.rt_init_col]].drop_duplicates()
+            rt_init_profile = rt_init_profile.dropna()
+            rt_init_profile = torch.tensor(rt_init_profile[self.rt_init_col].values).unsqueeze(-1).to(torch.float32)
+            rt_init_profile = self.convert_rt_prior_units(rt_init_profile)
+            assert cn_s_reads.shape[0] == gammas.shape[0] == rt_init_profile.shape[0]
+        else:
+            rt_init_profile = None
 
-        return cn_g1_reads_df, cn_g1_states_df, cn_s_reads_df, cn_s_states_df, cn_g1_reads, cn_g1_states, cn_s_reads, cn_s_states, gammas, rt_prior_profile, libs_g1, libs_s
+        return cn_g1_reads_df, cn_g1_states_df, cn_s_reads_df, cn_s_states_df, cn_g1_reads, cn_g1_states, cn_s_reads, cn_s_states, gammas, rt_prior_profile, rt_init_profile, libs_g1, libs_s
 
 
     def sort_by_cell_and_loci(self, cn):
@@ -655,7 +665,7 @@ def run_pert_model(self):
 
         cn_g1_reads_df, cn_g1_states_df, cn_s_reads_df, cn_s_states_df, \
             cn_g1_reads, cn_g1_states, cn_s_reads, cn_s_states, \
-            gammas, rt_prior_profile, libs_g1, libs_s = self.process_input_data()
+            gammas, rt_prior_profile, rho_init, libs_g1, libs_s = self.process_input_data()
 
         # compute consensus clone profiles for cn state
         clone_cn_profiles = compute_consensus_clone_profiles(
@@ -779,12 +789,6 @@ def run_pert_model(self):
         # use manhattan binarization method to come up with an initial guess for each cell's time in S-phase
         t_init, t_alpha_prior, t_beta_prior = self.guess_times(cn_s_reads, etas)
 
-        # if the user has provided an initialization for RT, extract that as a tensor
-        if self.rt_init_col is not None:
-            rho_init = torch.tensor(cn_s_reads_df[self.rt_init_col].values)
-        else:
-            rho_init = None
-
         # condition gc betas of S-phase model using fitted results from G1-phase model
         model_s = poutine.condition(
             model_s,