Update curation pipeline

.gitignore

@@ -1 +1,3 @@
-*~
+*~
+data
+intermediate

src/00_setup_curation.sh

@@ -4,14 +4,11 @@
 
 # from a base directory
 
-
-mkdir data
-mkdir intermediate
-
 git clone https://<user_name>:<security_token>huggingface.co/RosettaCommons/MegaScale
 
+
 # needed to get splits
-git clone https://github.com/Kuhlman-Lab/ThermoMPNN.git
+cd data && git clone https://github.com/Kuhlman-Lab/ThermoMPNN.git
 
 
 

src/02.1_gather_ThermoMPNN_splits.py

@@ -0,0 +1,71 @@
+
+
+import pandas
+import pyarrow.parquet
+import pickle
+
+
+
+
+
+# Gabe requested that the splits defined in ThermoMPNN of the MegaScale dataset be the default splits
+
+# ThermoMPNN/datsets.py
+#   class MegaScaleDataset
+#     def __init__(csg, split):
+#         fname = self.cfg.data_loc.megascale_csv
+#         df = pd.read_csv(fname, usecols=["ddG_ML", "mut_type", "WT_name", "aa_seq", "dG_ML"])
+#
+#         # remove unreliable data and more complicated mutations
+#         df = df.loc[df.ddG_ML != '-', :].reset_index(drop=True)
+#         df = df.loc[
+#            ~df.mut_type.str.contains("ins") &
+#            ~df.mut_type.str.contains("del") &
+#            ~df.mut_type.str.contains(":"), :].reset_index(drop=True)
+#
+#         splits = <load from self.cfg.data_loc.megascale_splits>
+# 
+#         if self.split != 'all' and (cfg.reduce != 'prot' or self.split != 'train'):
+#           self.wt_names = splits[split]
+#
+#
+#
+
+# local.yaml
+#   data_loc:
+#     megascale_csv: "<truncated>/Processed_K50_dG_datasets/Tsuboyama2023_Dataset2_Dataset3_20230416.csv"
+
+
+
+
+
+
+
+
+with open("data/ThermoMPNN/dataset_splits/mega_splits.pkl", "rb") as f:
+    mega_splits  = pickle.load(f)
+
+splits = []
+for split_name, split_ids in mega_splits.items():
+    splits.append(
+        pandas.DataFrame({
+            'split_name': split_name,
+            'id': split_ids}))
+
+splits = pandas.concat(splits)
+splits.reset_index(drop=True, inplace=True)
+
+pyarrow.parquet.write_table(
+    pyarrow.Table.from_pandas(splits),
+    where = "intermediate/ThermoMPNN_splits.parquet")
+
+
+parquet_file = pyarrow.parquet.ParquetFile('intermediate/ThermoMPNN_splits.parquet')
+parquet_file.metadata
+# <pyarrow._parquet.FileMetaData object at 0x149f5d2667a0>
+#   created_by: parquet-cpp-arrow version 17.0.0
+#   num_columns: 2
+#   num_rows: 2020
+#   num_row_groups: 1
+#   format_version: 2.6
+#   serialized_size: 1881

src/{02.1_assemble_K50_dG_dataset.R → 02.2_assemble_K50_dG_dataset.R}

@@ -1,7 +1,14 @@
+
+
 system("cd data; unzip Processed_K50_dG_datasets.zip")
 
 
+ThermoMPNN_splits <- arrow::read_parquet("intermediate/ThermoMPNN_splits.parquet")
+
+
 ### Dataset1 ###
+
+# Dataset1 consists of all cDNA proteolysis measurements of stability
 dataset1 <- readr::read_csv(
     file = "data/Processed_K50_dG_datasets/Tsuboyama2023_Dataset1_20230416.csv",
     col_types = readr::cols(
@@ -43,8 +50,65 @@ dataset1 |>
 
 
 
+### Dataset2 and Dataset3 ###
+
+# Dataset2 (for dG ML) consists of cDNA proteolysis measurements of stability that are of class G0 + G1
+# Datase3 (for ddG ML) consists of cDNA proteolysis measurements of stability that are of class G0
+#    G0: Good (wild-type ΔG values below 4.75 kcal mol^−1), 325,132 ΔG measurements at 17,093 sites in 365 domains
+#    G1: Good but WT outside dynamic range
+
 dataset23 <- readr::read_csv(
-    file = "data/Processed_K50_dG_datasets/Tsuboyama2023_Dataset2_Dataset3_20230416.csv"
+    file = "data/Processed_K50_dG_datasets/Tsuboyama2023_Dataset2_Dataset3_20230416.csv",
+    show_col_types = FALSE)
+# 776,298 rows
+
+dataset23 |>
+    arrow::write_parquet(
+        "intermediate/dataset23.parquet")
+
+dataset3 <- dataset23 |>
+    dplyr::filter(ddG_ML != "-")
+
+dataset3_single_mutant <- dataset3 |>
+    dplyr::filter(!(mut_type |> stringr::str_detect("(ins|del|[:])")))
+        
+
+ThermoMPNN_splits |> dplyr::group_by(split_name) |>
+    dplyr::do({
+        split <- .
+        split_name <- split$split_name[1]
+        mutant_set <- dataset3_single_mutant |>
+            dplyr::filter(mut_type != "wt") |>
+            dplyr::semi_join(split, by = c("WT_name" = "id"))
+        cat("Writing out split ", split_name, ", nrow: ", nrow(mutant_set), "\n", sep = "")
+        
+        arrow::write_parquet(
+            x = mutant_set,
+            sink = paste0("intermediate/dataset3_ThermoMPNN_", split_name, ".parquet"))
+        data.frame()
+    })
+        
+
+dataset3_single_mutant_train <- dataset3_single_mutant |>
+    dplyr::filter(mut_type != "wt") |>
+    dplyr::semi_join(
+        ThermoMPNN_splits |>
+        dplyr::filter(split_name == "train"),
+        by = c("WT_name" = "id"))
+
+dataset3_single_mutant_val <- dataset3_single_mutant |>
+    dplyr::filter(mut_type != "wt") |>
+    dplyr::semi_join(
+        ThermoMPNN_splits |>
+        dplyr::filter(split_name == "val"),
+        by = c("WT_name" = "id"))
+
+dataset3_single_mutant_test <- dataset3_single_mutant |>
+    dplyr::filter(mut_type != "wt") |>
+    dplyr::semi_join(
+        ThermoMPNN_splits |>
+        dplyr::filter(split_name == "test"),
+        by = c("WT_name" = "id"))
 
 
 ####

src/02.8_gather_ThermoMPNN_splits.py

@@ -1,20 +0,0 @@
-
-
-import pandas
-import pyarrow.parquet
-import pickle
-
-with open("ThermoMPNN/dataset_splits/mega_splits.pkl", "rb") as f:
-    mega_splits  = pickle.load(f)
-
-splits = []
-for split_name, split_ids in mega_splits.items():
-    splits.append(
-        pandas.DataFrame({
-            'split_name': split_name,
-            'id': split_ids}))
-
-splits = pandas.concat(splits)
-pyarrow.parquet.write_table(
-    pyarrow.Table.from_pandas(splits),
-    where = "intermediate/ThermoMPNN_splits.parquet")

src/03.1_upload_data.py

@@ -0,0 +1,79 @@
+
+
+
+# install huggingface_hub from the command line:
+#
+#    pip install huggingface_hub
+#    pip install datasets
+#
+# Log into huggingface hub (this only needs to be done once per project, and then it is cached)
+#  
+#    huggingface-cli login
+# 
+# This will ask you for an access token
+
+
+import datasets
+
+
+# Dataset1
+# Dataset2
+# Dataset3
+#  Single Mutants
+#    
+
+# dataset1
+# dataset2
+# dataset3
+# dataset3_single
+# dataset3_single_CV
+
+
+
+
+##### dataset3_single #######
+dataset = datasets.load_dataset(
+    "parquet",
+    name = "dataset3_single",
+    data_dir = "./intermediate",
+    data_files = {
+        "train" : "dataset3_ThermoMPNN_train.parquet",
+        "val" : "dataset3_ThermoMPNN_val.parquet",
+        "test" : "dataset3_ThermoMPNN_test.parquet"},
+    cache_dir = "/scratch/maom_root/maom0/maom",
+    keep_in_memory = True)
+
+dataset.push_to_hub(
+    repo_id = "maom/MegaScale",
+    config_name = "dataset3_single",
+    data_dir = "dataset3_single/data")
+
+
+##### dataset3_single #######
+dataset = datasets.load_dataset(
+    "parquet",
+    name = "dataset3_single_CV",
+    data_dir = "./intermediate",
+    data_files = {
+        "train_0" : "dataset3_ThermoMPNN_train_0.parquet",
+        "train_1" : "dataset3_ThermoMPNN_train_1.parquet",        
+        "train_2" : "dataset3_ThermoMPNN_train_2.parquet",
+        "train_3" : "dataset3_ThermoMPNN_train_3.parquet",
+        "train_4" : "dataset3_ThermoMPNN_train_4.parquet",        
+        "val_0" : "dataset3_ThermoMPNN_val_0.parquet",
+        "val_1" : "dataset3_ThermoMPNN_val_1.parquet",        
+        "val_2" : "dataset3_ThermoMPNN_val_2.parquet",
+        "val_3" : "dataset3_ThermoMPNN_val_3.parquet",
+        "val_4" : "dataset3_ThermoMPNN_val_4.parquet",        
+        "test_0" : "dataset3_ThermoMPNN_test_0.parquet",
+        "test_1" : "dataset3_ThermoMPNN_test_1.parquet",        
+        "test_2" : "dataset3_ThermoMPNN_test_2.parquet",
+        "test_3" : "dataset3_ThermoMPNN_test_3.parquet",
+        "test_4" : "dataset3_ThermoMPNN_test_4.parquet"},
+    cache_dir = "/scratch/maom_root/maom0/maom",
+    keep_in_memory = True)
+
+dataset.push_to_hub(
+    repo_id = "MaomLab/MegaScale",
+    config_name = "dataset3_single_CV",
+    data_dir = "datase3_single/data")