File size: 3,170 Bytes
998df90
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b39b124
998df90
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
---
license: mit
tags:
- molecular-property-prediction
- graph-neural-network
- chemistry
- pytorch
datasets:
- qm9
- spice
- pfas
metrics:
- mse
- mae
pipeline_tag: graph-ml
library_name: moml
---

# djmgnn-base

## Model Description

This is a base **DJMGNN** (Dense Jump Multi-Graph Neural Network) model for molecular property prediction. The model is designed to predict various molecular properties from graph representations of molecules.

### Architecture

- **Model Type**: Dense Jump Multi-Graph Neural Network (DJMGNN)
- **Framework**: PyTorch
- **Library**: MoML (Molecular Machine Learning)
- **Task**: Molecular Property Prediction

### Model Architecture Details

- **Hidden Dimensions**: 128
- **Number of Blocks**: 3
- **Layers per Block**: 6
- **Input Node Dimensions**: 11
- **Input Edge Dimensions**: 0
- **Node Output Dimensions**: 3
- **Graph Output Dimensions**: 19
- **Energy Output Dimensions**: 1
- **Jumping Knowledge Mode**: cat
- **Dropout Rate**: 0.2
- **Uses Supernode**: True
- **Uses RBF Features**: True
- **RBF K**: 32

## Training Details

### Datasets

The model was trained on the following datasets:
- **QM9**: Quantum mechanical properties of small molecules
- **SPICE**: Molecular dynamics data with forces and energies
- **PFAS**: Per- and polyfluoroalkyl substances dataset

### Training Configuration

```yaml
batch_size: 32
early_stopping: true
epochs: 100
learning_rate: 0.001
optimizer: Adam
patience: 10
validation_split: 0.2

```

## Usage

### Loading the Model

```python
import torch
from moml.models.mgnn.djmgnn import DJMGNN

# Load the model
model = DJMGNN(
    in_node_dim=11,
    in_edge_dim=0,
    hidden_dim=128,
    n_blocks=3,
    layers_per_block=6,
    node_output_dims=3,
    graph_output_dims=19,
    energy_output_dims=1,
    jk_mode="cat",
    dropout=0.2,
    use_supernode=true,
    use_rbf=true,
    rbf_K=32
)

# Load the checkpoint
checkpoint = torch.load("path/to/pytorch_model.pt", map_location="cpu")
model.load_state_dict(checkpoint["model_state_dict"])
model.eval()
```

### Making Predictions

```python
# Assuming you have a molecular graph 'data' (torch_geometric.data.Data object)
with torch.no_grad():
    output = model(
        x=data.x,
        edge_index=data.edge_index,
        edge_attr=data.edge_attr,
        batch=data.batch
    )
    
    # Extract predictions
    node_predictions = output["node_pred"]      # Node-level predictions
    graph_predictions = output["graph_pred"]    # Graph-level predictions
    energy_predictions = output["energy_pred"]  # Energy predictions
```

## Model Performance

This is the base DJMGNN model trained on QM9, SPICE, and PFAS datasets.

## Citation

If you use this model in your research, please cite:

```bibtex
@misc{djmgnn_model,
  title={DJMGNN: Dense Jump Multi-Graph Neural Network for Molecular Property Prediction},
  author={Your Name},
  year={2024},
  url={https://github.com/SAKETH11111/MoML-CA}
}
```

## License

This model is released under the MIT License.

## Contact

For questions or issues, please contact sakethbaddam10@gmail.com or open an issue in the [GitHub repository](https://github.com/SAKETH11111/MoML-CA).