Complete equivariant embedding net

MakeGraph.py

@@ -3,8 +3,8 @@
 # ==================                                            # 
 # 1. Find RMSD of docked ligand poses from crystal structure    #
 # 2. Retrieve molecular features of protein and ligand          #
-# 3. Prepare node and edge features (ODDT)                      #
-# 4. Dump all relevant info to pickle file                      #
+# 3. Prepare node and edge features (ODDT/RDKit)                #
+# 4. Dump all relevant info to PyTorch .pt file                 #
 #                                                               #
 #################################################################
 
@@ -42,7 +42,7 @@
     parser.add_argument('--config', type=str,
                         default='./config/config.yml')
     parser.add_argument('--outdir', type=str, 
-                        default='./dataset/crossdocked_graph10')
+                        default='./dataset/crossdocked_graph10_v2')
     args = parser.parse_args()
 
     # Logging

README.md

@@ -10,11 +10,35 @@ Principal Investigator: **[Assoc Prof MU Yuguang](https://dr.ntu.edu.sg/cris/rp/
 
 ## Introduction
 
-![Protein-ligand complex as three-dimensional heterogeneous graph](img/PLComplex.png)
+![Protein-ligand complex as three-dimensional heterogeneous graph](img/PLGraph.png)
 <p align="center">
     <em>Protein-ligand complex as a three-dimensional heterogeneous graph, with an emphasis on interactions between the protein and ligand.</em>
 </p>
 
+## Installation
+
+Working inside a Conda virtual environment is hightly encouraged, but not necessary. Create a new environment with the command:
+
+```
+conda env create -f environment.yml
+```
+
+Otherwise, install the following dependencies sequentially.
+
+```
+conda create -n SINGA python=3.10
+conda activate SINGA
+conda install -c conda-forge openbabel
+pip install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
+pip install torch_geometric==2.3.1
+pip install torch-scatter==2.1.1 torch-sparse==0.6.17
+pip install biopandas==0.4.1 pytorch-lightning==2.0.6
+pip install rdkit==2023.3.3
+pip install oddt==0.7
+pip install easydict==1.10
+pip install e3nn==0.5.1
+```
+
 ## Directory tree
 
 ```
@@ -25,7 +49,7 @@ Principal Investigator: **[Assoc Prof MU Yuguang](https://dr.ntu.edu.sg/cris/rp/
   |__ /ckpt
   |__ /config
   |__ /dataset
-      |__ /crossdocked_graph10
+      |__ /crossdocked_graph10_v2
   |__ /example
   |__ /features
   |__ /img

config/train.yml

@@ -1,30 +1,37 @@
 dataset:
-    path: ./dataset/crossdocked_graph10
+    path: ./dataset/crossdocked_graph10_v2
     split: ./dataset/split_by_name.pt
+    atomic_distance_cutoff: 4.0
 
-embedding:
+dataloader:
     atomic_distance_cutoff: 4.0
     batch_size: 64
-    dropout: 0.13
-    heads: 2
-    hetero_aggr: mean
-    hidden_channels_la: 
-    - 48
-    - 76
-    - 121
-    hidden_channels_pa:
-    - 48
-    - 57
-    - 68
-    lr: 0.001
-    mlp_channels:
-    - 1
-    - 189
-    - 256
     molecular_embedding_size: 4
-    num_layers: 3
-    str_for_hparams: 'InterMol length: 4.0A'
-    weight_decay: 0.0001
+    num_workers: 1
 
+embedding:
+    edge_channels: 16
+    sphere_channels: 16
+    attn_hidden_channels: 128
+    attn_alpha_channels: 32
+    attn_value_channels: 16
+    ffn_hidden_channels: 512
+    lmax_list:
+    - 6
+    mmax_list:
+    - 2
+    cutoff: 10.0
+    max_num_elements: 43
+    num_heads: 7
+    num_layers: 3
+    norm_type: 'rms_norm_sh'
+    activation_type: 'scaled_silu'
+    use_atom_edge_embedding: True
+    share_atom_edge_embedding: True
+    grid_resolution: None
+    alpha_drop: 0.0
+    proj_drop: 0.0
+    drop_path_rate: 0.0
+
 model:
     name: SINGA