jyz-mbzuai commited on
Commit
648f3e9
1 Parent(s): 51a9333

update readme

Browse files
README.md CHANGED
@@ -14,12 +14,12 @@ AIDO.StructureTokenizer is a VQ-VAE-based tokenizer designed for protein structu
14
 
15
  ## Model Description
16
 
17
- TODO model figure
18
 
19
  **AIDO.StructureTokenizer** is built on a Vector Quantized Variational Autoencoder (VQ-VAE) architecture with the following components:
20
- - Equivariant Encoder: Encodes backbone structures into a latent space that maintains rotational and translational symmetries using the Equiformer architecture.
21
  - Discrete Codebook: Maps continuous latent vectors into 512 discrete structural tokens.
22
- - Invariant Decoder: Reconstructs full 3D structures, including side chains, from the structural tokens using an architecture adapted from ESMFold.
23
 
24
  This model strikes a balance between reconstruction fidelity and structural locality, optimizing its suitability for downstream tasks such as structure prediction, homology detection, and multimodal protein modeling.
25
 
@@ -30,17 +30,14 @@ This model strikes a balance between reconstruction fidelity and structural loca
30
  - Reconstructing Structures (See [below](#reconstructing-structures))
31
  - Structure Prediction (See [this section](https://huggingface.co/genbio-ai/AIDO.Protein2StructureToken-16B/blob/main/README.md#structure-prediction) in genbio-ai/AIDO.Protein2StructureToken-16B)
32
 
33
- ### Hyperparameters
34
-
35
- TODO
36
-
37
- ### Training details
38
-
39
- TODO
40
-
41
  ## Results
42
 
43
- TODO
 
 
 
 
 
44
 
45
  ## How to Use
46
  Please see `experiments/AIDO.StructureTokenizer` in [Model Generator](https://github.com/genbio-ai/modelgenerator) for more details.
 
14
 
15
  ## Model Description
16
 
17
+ ![Model Architecture](./assets/images/architecture.png)
18
 
19
  **AIDO.StructureTokenizer** is built on a Vector Quantized Variational Autoencoder (VQ-VAE) architecture with the following components:
20
+ - Equivariant Encoder (6M): Encodes backbone structures into a latent space that maintains rotational and translational symmetries using the Equiformer architecture.
21
  - Discrete Codebook: Maps continuous latent vectors into 512 discrete structural tokens.
22
+ - Invariant Decoder (300M): Reconstructs full 3D structures, including side chains, from the structural tokens using an architecture adapted from ESMFold.
23
 
24
  This model strikes a balance between reconstruction fidelity and structural locality, optimizing its suitability for downstream tasks such as structure prediction, homology detection, and multimodal protein modeling.
25
 
 
30
  - Reconstructing Structures (See [below](#reconstructing-structures))
31
  - Structure Prediction (See [this section](https://huggingface.co/genbio-ai/AIDO.Protein2StructureToken-16B/blob/main/README.md#structure-prediction) in genbio-ai/AIDO.Protein2StructureToken-16B)
32
 
 
 
 
 
 
 
 
 
33
  ## Results
34
 
35
+ ### Reconstructing Structures
36
+ ![Reconstruction Results](./assets/images/reconstruction.png)
37
+ ### Homology Detection
38
+ ![Homology Detection Results](./assets/images/homology_detection.png)
39
+ ### Structure Prediction
40
+ ![Structure Prediction Results](./assets/images/structure_prediction.png)
41
 
42
  ## How to Use
43
  Please see `experiments/AIDO.StructureTokenizer` in [Model Generator](https://github.com/genbio-ai/modelgenerator) for more details.
assets/images/architecture.png ADDED
assets/images/homology_detection.png ADDED
assets/images/reconstruction.png ADDED
assets/images/structure_prediction.png ADDED
assets/images/structure_tokenizer/launch_protein_viewer.png ADDED
assets/images/structure_tokenizer/select_files.png ADDED
assets/images/structure_tokenizer/visualize_reconstruction.png ADDED