Update pre-trained DeepRVAT models (#91)

* update pretrained model with 30 models (5 cv folds , each with 6 repeats) * remove out-dated checkpoint files * specifying new pretrained model path * reorganize directories for pipeline running from pretrained models * add in missing alphamissence to annotations * adding in missing alphamissense annotation * specifying random floats between 0 and 1 for example data in alphamissense annotation
PMBio · May 21, 2024 · 8bddce6 · 8bddce6
1 parent 82cc76f
commit 8bddce6
Show file tree

Hide file tree

Showing 44 changed files with 31 additions and 23 deletions.
diff --git a/example/annotations.parquet b/example/annotations.parquet
diff --git a/example/config.yaml b/example/config.yaml
@@ -158,6 +158,7 @@ training_data:
             - DeepRipe_plus_MBNL1_parclip
             - DeepRipe_plus_QKI_parclip
             - SpliceAI_delta_score
+            - alphamissense
         use_common_variants: False
         use_rare_variants: True
         rare_embedding:
@@ -197,6 +198,7 @@ training_data:
                     - DeepRipe_plus_MBNL1_parclip
                     - DeepRipe_plus_QKI_parclip
                     - SpliceAI_delta_score
+                    - alphamissense
                 thresholds:
                     MAF: "MAF < 1e-2"
                     CADD_PHRED:  "CADD_PHRED > 5"
@@ -277,6 +279,7 @@ data:
             - DeepRipe_plus_MBNL1_parclip
             - DeepRipe_plus_QKI_parclip
             - SpliceAI_delta_score
+            - alphamissense
         gene_file: protein_coding_genes.parquet
         use_common_variants: False
         use_rare_variants: True
@@ -317,6 +320,7 @@ data:
                     - DeepRipe_plus_MBNL1_parclip
                     - DeepRipe_plus_QKI_parclip
                     - SpliceAI_delta_score
+                    - alphamissense
                 thresholds:
                     MAF: "MAF < 1e-3"
                     CADD_PHRED:  "CADD_PHRED > 5"

diff --git a/pretrained_models/config.yaml b/pretrained_models/config.yaml
@@ -1,24 +1,24 @@
 model:
-    type: DeepSet
-    model_collection: agg_models
-    checkpoint: combined_agg.pt
-    config:
-        phi_layers: 2
-        phi_hidden_dim: 20
-        rho_layers: 3
-        rho_hidden_dim: 10
-        activation: LeakyReLU
-        pool: max
-        use_sigmoid: True
-        metrics:
-            objective: MSE
-            objective_mode: min
-            loss: MSE
-            all:
-                MSE: {}
-                PearsonCorrTorch: {}
-                MAE: {}
-                RSquared: {}
-        optimizer:
-            type: AdamW
-            config: {}
+  checkpoint: combined_agg.pt
+  config:
+    activation: LeakyReLU
+    metrics:
+      all:
+        MAE: {}
+        MSE: {}
+        PearsonCorrTorch: {}
+        RSquared: {}
+      loss: MSE
+      objective: MSE
+      objective_mode: min
+    optimizer:
+      config: {}
+      type: AdamW
+    phi_hidden_dim: 20
+    phi_layers: 2
+    pool: max
+    rho_hidden_dim: 10
+    rho_layers: 3
+    use_sigmoid: true
+  model_collection: agg_models
+  type: DeepSet
diff --git a/pretrained_models/repeat_0/best/bag_0.ckpt b/pretrained_models/repeat_0/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_1/best/bag_0.ckpt b/pretrained_models/repeat_1/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_1/best/bag_0.ckpt.reverse b/pretrained_models/repeat_1/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_10/best/bag_0.ckpt b/pretrained_models/repeat_10/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_11/best/bag_0.ckpt b/pretrained_models/repeat_11/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_12/best/bag_0.ckpt b/pretrained_models/repeat_12/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_12/best/bag_0.ckpt.reverse b/pretrained_models/repeat_12/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_13/best/bag_0.ckpt b/pretrained_models/repeat_13/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_14/best/bag_0.ckpt b/pretrained_models/repeat_14/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_15/best/bag_0.ckpt b/pretrained_models/repeat_15/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_16/best/bag_0.ckpt b/pretrained_models/repeat_16/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_16/best/bag_0.ckpt.reverse b/pretrained_models/repeat_16/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_17/best/bag_0.ckpt b/pretrained_models/repeat_17/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_18/best/bag_0.ckpt b/pretrained_models/repeat_18/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_19/best/bag_0.ckpt b/pretrained_models/repeat_19/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_2/best/bag_0.ckpt b/pretrained_models/repeat_2/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_20/best/bag_0.ckpt b/pretrained_models/repeat_20/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_20/best/bag_0.ckpt.reverse b/pretrained_models/repeat_20/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_21/best/bag_0.ckpt b/pretrained_models/repeat_21/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_22/best/bag_0.ckpt b/pretrained_models/repeat_22/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_23/best/bag_0.ckpt b/pretrained_models/repeat_23/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_23/best/bag_0.ckpt.reverse b/pretrained_models/repeat_23/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_24/best/bag_0.ckpt b/pretrained_models/repeat_24/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_25/best/bag_0.ckpt b/pretrained_models/repeat_25/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_25/best/bag_0.ckpt.reverse b/pretrained_models/repeat_25/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_26/best/bag_0.ckpt b/pretrained_models/repeat_26/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_27/best/bag_0.ckpt b/pretrained_models/repeat_27/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_27/best/bag_0.ckpt.reverse b/pretrained_models/repeat_27/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_28/best/bag_0.ckpt b/pretrained_models/repeat_28/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_29/best/bag_0.ckpt b/pretrained_models/repeat_29/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_3/best/bag_0.ckpt b/pretrained_models/repeat_3/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_4/best/bag_0.ckpt b/pretrained_models/repeat_4/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_5/best/bag_0.ckpt b/pretrained_models/repeat_5/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_5/best/bag_0.ckpt.reverse b/pretrained_models/repeat_5/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_6/best/bag_0.ckpt b/pretrained_models/repeat_6/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_6/best/bag_0.ckpt.reverse b/pretrained_models/repeat_6/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_7/best/bag_0.ckpt b/pretrained_models/repeat_7/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_7/best/bag_0.ckpt.reverse b/pretrained_models/repeat_7/best/bag_0.ckpt.reverse
diff --git a/pretrained_models/repeat_8/best/bag_0.ckpt b/pretrained_models/repeat_8/best/bag_0.ckpt
diff --git a/pretrained_models/repeat_9/best/bag_0.ckpt b/pretrained_models/repeat_9/best/bag_0.ckpt
diff --git a/tests/deeprvat/test_data/training/config.yaml b/tests/deeprvat/test_data/training/config.yaml
@@ -194,6 +194,7 @@ training_data:
             - DeepRipe_plus_MBNL1_parclip
             - DeepRipe_plus_QKI_parclip
             - SpliceAI_delta_score
+            - alphamissense
         use_common_variants: False
         use_rare_variants: True
         verbose: True
@@ -234,6 +235,7 @@ training_data:
                     - DeepRipe_plus_MBNL1_parclip
                     - DeepRipe_plus_QKI_parclip
                     - SpliceAI_delta_score
+                    - alphamissense
                 thresholds:
                     MAF: "MAF < 1e-2"
                     CADD_PHRED:  "CADD_PHRED > 5"
@@ -312,6 +314,7 @@ data:
             - DeepRipe_plus_MBNL1_parclip
             - DeepRipe_plus_QKI_parclip
             - SpliceAI_delta_score
+            - alphamissense
         gene_file: example/protein_coding_genes.parquet
         use_common_variants: False
         use_rare_variants: True
@@ -352,6 +355,7 @@ data:
                     - DeepRipe_plus_MBNL1_parclip
                     - DeepRipe_plus_QKI_parclip
                     - SpliceAI_delta_score
+                    - alphamissense
                 thresholds:
                     MAF: "MAF < 1e-3"
                     CADD_PHRED:  "CADD_PHRED > 5"