Delete README.md
Browse files
README.md
DELETED
|
@@ -1,70 +0,0 @@
|
|
| 1 |
-
## Introduction
|
| 2 |
-
<p align="center">
|
| 3 |
-
<br>
|
| 4 |
-
<img src="assets/FAPM.png"/>
|
| 5 |
-
<br>
|
| 6 |
-
<p>
|
| 7 |
-
|
| 8 |
-
## Installation
|
| 9 |
-
|
| 10 |
-
1. (Optional) Creating conda environment
|
| 11 |
-
|
| 12 |
-
```bash
|
| 13 |
-
conda create -n lavis python=3.8
|
| 14 |
-
conda activate lavis
|
| 15 |
-
```
|
| 16 |
-
|
| 17 |
-
2. for development, you may build from source
|
| 18 |
-
|
| 19 |
-
```bash
|
| 20 |
-
git clone https://github.com/xiangwenkai/FAPM.git
|
| 21 |
-
cd FAPM
|
| 22 |
-
pip install -e .
|
| 23 |
-
|
| 24 |
-
pip install Biopython
|
| 25 |
-
pip install fair-esm
|
| 26 |
-
```
|
| 27 |
-
|
| 28 |
-
### Datasets
|
| 29 |
-
#### 1.raw dataset
|
| 30 |
-
Raw data are avaliable at *https://ftp.uniprot.org/pub/databases/uniprot/previous_releases/release-2023_04/knowledgebase/*, this file is very large and need to be processed to get its name, sequence, GO label, function description and prompt.
|
| 31 |
-
The domain level protein dataset we used are avaliable at *https://ftp.ebi.ac.uk/pub/databases/interpro/releases/95.0/protein2ipr.dat.gz*
|
| 32 |
-
In this respository, We provide the experimental train/val/test sets of Swiss-Prot, which are avaliable at data/swissprot_exp
|
| 33 |
-
#### 2.ESM2 embeddings
|
| 34 |
-
ESM2 embeddings generation code: *https://github.com/facebookresearch/esm*
|
| 35 |
-
The generation command:
|
| 36 |
-
```bash
|
| 37 |
-
git clone https://github.com/facebookresearch/esm.git
|
| 38 |
-
python scripts/extract.py esm2_t33_3B_UR50D you_path/protein.fasta you_path_to_save_embedding_files --repr_layers 36 --truncation_seq_length 1024 --include per_tok
|
| 39 |
-
```
|
| 40 |
-
The default path to save embedding files in this respository is **data/emb_esm2_3b**
|
| 41 |
-
|
| 42 |
-
## Pretraining language models
|
| 43 |
-
Source: *https://huggingface.co/teknium/OpenHermes-2.5-Mistral-7B*
|
| 44 |
-
|
| 45 |
-
## Training
|
| 46 |
-
data config: lavis/configs/datasets/protein/GO_defaults_cap.yaml
|
| 47 |
-
stage1 config: lavis/projects/blip2/train/protein_pretrain_stage1.yaml
|
| 48 |
-
stage1 training command: run_scripts/blip2/train/protein_pretrain_domain_stage1.sh
|
| 49 |
-
stage2 config: lavis/projects/blip2/train/protein_pretrain_stage2.yaml
|
| 50 |
-
stage2 training/finetuning command: run_scripts/blip2/train/protein_pretrain_domain_stage2.sh
|
| 51 |
-
|
| 52 |
-
## Trained models
|
| 53 |
-
You can download our trained models from drive: *https://drive.google.com/drive/folders/1aA0eSYxNw3DvrU5GU1Cu-4q2kIxxAGSE?usp=drive_link*
|
| 54 |
-
|
| 55 |
-
## Testing
|
| 56 |
-
config: lavis/projects/blip2/eval/caption_protein_eval.yaml
|
| 57 |
-
command: run_scripts/blip2/eval/eval_cap_protein.sh
|
| 58 |
-
|
| 59 |
-
## Inference example
|
| 60 |
-
We provide an example in **FAPM_inference.py**. You can change the example protein to you custom case
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|