Code improvements to run on subsets of the original dataset

configs/lgi/all.yaml

@@ -0,0 +1,32 @@
+seed: 42
+#root_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_data
+#logs_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_logs
+#origin: /home/daniel/Desktop/GIFFLAR/lgi_data_full.pkl
+root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/lgi_data
+logs_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/lgi_logs
+origin: /home/rjo21/Desktop/GIFFLAR/lgi_data_20.pkl
+model:
+  glycan_encoder:
+    - name: gifflar
+      feat_dim: 128
+      hidden_dim: 1024
+      num_layers: 8
+      pooling: global_mean
+    - name: sweetnet
+      feat_dim: 128
+      hidden_dim: 1024
+      num_layers: 16
+  lectin_encoder:
+    - name: ESM
+      layer_num: 33
+    - name: Ankh
+      layer_num: 48
+    - name: ProtBert
+      layer_num: 30
+    - name: ProstT5
+      layer_num: 24
+  batch_size: 256
+  epochs: 100
+  learning_rate: 0.001
+  optimizer: Adam
+

configs/lgi/full.yaml

@@ -1,7 +1,10 @@
 seed: 42
-root_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_data
-logs_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_logs
-origin: /home/daniel/Desktop/GIFFLAR/lgi_data_full.pkl
+#root_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_data
+#logs_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_logs
+#origin: /home/daniel/Desktop/GIFFLAR/lgi_data_full.pkl
+root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/lgi_data
+logs_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/lgi_logs
+origin: /home/rjo21/Desktop/GIFFLAR/lgi_data_20.pkl
 model:
   glycan_encoder:
     name: gifflar

configs/lgi/test.yaml

@@ -1,7 +1,10 @@
 seed: 42
-root_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_data
-logs_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_logs
-origin: /home/daniel/Desktop/GIFFLAR/lgi_data.pkl
+#root_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_data
+#logs_dir: /home/daniel/Data1/roman/GIFFLAR/lgi_logs
+#origin: /home/daniel/Desktop/GIFFLAR/lgi_data_full.pkl
+root_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/lgi_data
+logs_dir: /scratch/SCRATCH_SAS/roman/Gothenburg/GIFFLAR/lgi_logs
+origin: /home/rjo21/Desktop/GIFFLAR/lgi_data.pkl
 model:
   glycan_encoder:
     name: gifflar

experiments/aquire_lgi_dataset.py

@@ -1,4 +1,5 @@
 import pickle
+import random
 
 from tqdm import tqdm
 import numpy as np
@@ -22,6 +23,8 @@
 for i, ((aa_seq, iupac), val) in tqdm(enumerate(s.items())):
     data.append((lectins[aa_seq], glycans[iupac], val, splits[i]))
 
-with open("lgi_data_full.pkl", "wb") as f:
+data = random.sample(data, int(len(data) * 0.20))
+
+with open("lgi_data_20.pkl", "wb") as f:
     pickle.dump((data, lectins, glycans), f)
 

experiments/train_lgi.py

@@ -1,5 +1,5 @@
 import os
-os.environ["CUDA_VISIBLE_DEVICES"] = "1"
+# os.environ["CUDA_VISIBLE_DEVICES"] = "1"
 
 from argparse import ArgumentParser
 import time
@@ -19,15 +19,37 @@
 from gifflar.train import setup
 from gifflar.utils import read_yaml_config, hash_dict
 
+import torch
+print(torch.cuda.is_available())
 
 GLYCAN_ENCODERS = {
     "gifflar": GlycanGIN,
     "sweetnet": SweetNetLightning,
 }
 
 
-def main(config):
-    kwargs = read_yaml_config(config)
+def unfold_config(config: dict):
+    if isinstance(config["model"]["glycan_encoder"], dict):
+        ges = [config["model"]["glycan_encoder"]]
+    else:
+        ges = config["model"]["glycan_encoder"]
+    del config["model"]["glycan_encoder"]
+
+    if isinstance(config["model"]["lectin_encoder"], dict):
+        les = [config["model"]["lectin_encoder"]]
+    else:
+        les = config["model"]["lectin_encoder"]
+    del config["model"]["lectin_encoder"]
+
+    for le in les:
+        for ge in ges:
+            tmp_config = copy.deepcopy(config)
+            tmp_config["model"]["lectin_encoder"] = le
+            tmp_config["model"]["glycan_encoder"] = ge
+            yield tmp_config
+
+
+def train(**kwargs):
     kwargs["pre-transforms"] = {"GIFFLARTransform": "", "SweetNetTransform": ""}
     kwargs["hash"] = hash_dict(kwargs["pre-transforms"])
     seed_everything(kwargs["seed"])
@@ -63,6 +85,12 @@ def main(config):
     print("Training took", time.time() - start, "s")
 
 
+def main(config):
+    custom_args = read_yaml_config(config)
+    for args in unfold_config(custom_args):
+        train(**args)
+
+
 if __name__ == '__main__':
     parser = ArgumentParser()
     parser.add_argument("config", type=str, help="Path to YAML config file")

gifflar/train.py

@@ -1,5 +1,5 @@
 import os
-os.environ["CUDA_VISIBLE_DEVICES"] = "1"
+# os.environ["CUDA_VISIBLE_DEVICES"] = "1"
 
 from pathlib import Path
 from typing import Any

requirements.txt

@@ -4,12 +4,15 @@ rdkit>=2022
 scikit-learn
 numpy
 pandas
-glyles
-git+https://github.com/BojarLab/glycowork.git@1123e1cee9f189ea5e82c6c1cda9764749b6da36
+glyles>=1.0.0
+git+https://github.com/BojarLab/glycowork.git@145877da82cf0200b062945dfb85584b9d9ef30d
 jsonargparse
 rich
 pytorch-lightning
 pytest
 pyyaml
 networkx
 torchmetrics
+transformers
+sentencepiece
+xformers==0.0.28.post1