GitHub - ai4protein/VenusFactory: 🏭 AI agent platform for protein engineering. (ACL Demo 2025)

✏️ Table of Contents

Features
Supported Models
Supported Training Approaches
Supported Datasets
Supported Metrics
Agent-0.1(Beta Version)
Quick Tools。
Advanced Tools
Requirements
Installation Guide
Quick Start with Venus Web UI
Code-line Usage
Citation
Acknowledgement

📑 Features

🙌 VenusFactory is a unified open platform for protein engineering, supporting both graphical user interface (GUI) and command-line operations. It enables data retrieval, model training, evaluation, and deployment through a streamlined, no-code workflow.

🆒 With support for local private deployment and access to over 40 state-of-the-art deep learning models, VenusFactory lowers the barrier to scientific research and accelerates the application of AI in life sciences.

AI-Powered Assistance: Agent-0.1 acts as an intelligent AI assistant, providing expert answers and analysis for protein engineering tasks.
Efficient Workflows: Quick Tools enables rapid, no-code predictions for common tasks like protein function and mutation effect scoring.
Advanced Analysis: Advanced Tools offers powerful, in-depth analysis for both sequence-based and structure-based zero-shot mutation predictions.
Various protein language models: Venus series, ESM series, ProtTrans series, Ankh series, etc.
Comprehensive supervised datasets: Localization, Fitness, Solubility, Stability, etc.
Easy and quick data collector: AlphaFold2 Database, RCSB, InterPro, UniProt, etc.
Experiment monitors: Wandb, Local
Friendly interface: Gradio UI

🧬 Supported Models (Zero-shot prediction)

🤖 Supported Models (Training for supervised tasks)

Pre-training Protein Language Models

Venus Series Models (Published by Liang's Lab)

Model	Size	Parameters	GPU Memory	Features	Template
ProSST-20	20	110M	4GB+	Mutation	AI4Protein/ProSST-20
ProSST-128	128	110M	4GB+	Mutation	AI4Protein/ProSST-128
ProSST-512	512	110M	4GB+	Mutation	AI4Protein/ProSST-512
ProSST-1024	1024	110M	4GB+	Mutation	AI4Protein/ProSST-1024
ProSST-2048	2048	110M	4GB+	Mutation	AI4Protein/ProSST-2048
ProSST-4096	4096	110M	4GB+	Mutation	AI4Protein/ProSST-4096
ProPrime-690M	690M	690M	16GB+	OGT-prediction	AI4Protein/Prime_690M
ProPrime-650M-OGT	650M	650M	16GB+	OGT-prediction	AI4Protein/ProPrime-650M-OGT
VenusPLM-300M	300M	300M	12GB+	Protein-language	AI4Protein/VenusPLM-300M

💡 These models often excel in specific tasks or offer unique architectural benefits

Venus-PETA Models: Tokenization variants

BPE Tokenization Series

Model	Vocab Size	Parameters	GPU Memory	Template
PETA-base	base	80M	4GB+	AI4Protein/deep_base
PETA-bpe-50	50	80M	4GB+	AI4Protein/deep_bpe_50
PETA-bpe-100	100	80M	4GB+	AI4Protein/deep_bpe_100
PETA-bpe-200	200	80M	4GB+	AI4Protein/deep_bpe_200
PETA-bpe-400	400	80M	4GB+	AI4Protein/deep_bpe_400
PETA-bpe-800	800	80M	4GB+	AI4Protein/deep_bpe_800
PETA-bpe-1600	1600	80M	4GB+	AI4Protein/deep_bpe_1600
PETA-bpe-3200	3200	80M	4GB+	AI4Protein/deep_bpe_3200

Unigram Tokenization Series

Model	Vocab Size	Parameters	GPU Memory	Template
PETA-unigram-50	50	80M	4GB+	AI4Protein/deep_unigram_50
PETA-unigram-100	100	80M	4GB+	AI4Protein/deep_unigram_100
PETA-unigram-200	200	80M	4GB+	AI4Protein/deep_unigram_200
PETA-unigram-400	400	80M	4GB+	AI4Protein/deep_unigram_400
PETA-unigram-800	800	80M	4GB+	AI4Protein/deep_unigram_800
PETA-unigram-1600	1600	80M	4GB+	AI4Protein/deep_unigram_1600
PETA-unigram-3200	3200	80M	4GB+	AI4Protein/deep_unigram_3200

💡 Different tokenization strategies may be better suited for specific tasks

ESM Series Models: Meta AI's protein language models

Model	Size	Parameters	GPU Memory	Training Data	Template
ESM2-8M	8M	8M	2GB+	UR50/D	facebook/esm2_t6_8M_UR50D
ESM2-35M	35M	35M	4GB+	UR50/D	facebook/esm2_t12_35M_UR50D
ESM2-150M	150M	150M	8GB+	UR50/D	facebook/esm2_t30_150M_UR50D
ESM2-650M	650M	650M	16GB+	UR50/D	facebook/esm2_t33_650M_UR50D
ESM2-3B	3B	3B	24GB+	UR50/D	facebook/esm2_t36_3B_UR50D
ESM2-15B	15B	15B	40GB+	UR50/D	facebook/esm2_t48_15B_UR50D
ESM-1b	650M	650M	16GB+	UR50/S	facebook/esm1b_t33_650M_UR50S
ESM-1v-1	650M	650M	16GB+	UR90/S	facebook/esm1v_t33_650M_UR90S_1
ESM-1v-2	650M	650M	16GB+	UR90/S	facebook/esm1v_t33_650M_UR90S_2
ESM-1v-3	650M	650M	16GB+	UR90/S	facebook/esm1v_t33_650M_UR90S_3
ESM-1v-4	650M	650M	16GB+	UR90/S	facebook/esm1v_t33_650M_UR90S_4
ESM-1v-5	650M	650M	16GB+	UR90/S	facebook/esm1v_t33_650M_UR90S_5

💡 ESM2 models are the latest generation, offering better performance than ESM-1b/1v

BERT-based Models: Transformer encoder architecture

Model	Size	Parameters	GPU Memory	Training Data	Template
ProtBert-Uniref100	420M	420M	12GB+	UniRef100	Rostlab/prot_bert
ProtBert-BFD	420M	420M	12GB+	BFD100	Rostlab/prot_bert_bfd
IgBert	420M	420M	12GB+	Antibody	Exscientia/IgBert
IgBert-unpaired	420M	420M	12GB+	Antibody	Exscientia/IgBert_unpaired

💡 BFD-trained models generally show better performance on structure-related tasks

T5-based Models: Encoder-decoder architecture

Model	Size	Parameters	GPU Memory	Training Data	Template
ProtT5-XL-UniRef50	3B	3B	24GB+	UniRef50	Rostlab/prot_t5_xl_uniref50
ProtT5-XXL-UniRef50	11B	11B	40GB+	UniRef50	Rostlab/prot_t5_xxl_uniref50
ProtT5-XL-BFD	3B	3B	24GB+	BFD100	Rostlab/prot_t5_xl_bfd
ProtT5-XXL-BFD	11B	11B	40GB+	BFD100	Rostlab/prot_t5_xxl_bfd
IgT5	3B	3B	24GB+	Antibody	Exscientia/IgT5
IgT5-unpaired	3B	3B	24GB+	Antibody	Exscientia/IgT5_unpaired
Ankh-base	450M	450M	12GB+	Encoder-decoder	ElnaggarLab/ankh-base
Ankh-large	1.2B	1.2B	20GB+	Encoder-decoder	ElnaggarLab/ankh-large

💡 T5 models can be used for both encoding and generation tasks

Model Selection Guide

How to choose the right model?

Based on Hardware Constraints:
- Limited GPU (<8GB): ESM2-8M, ESM2-35M, ProSST
- Medium GPU (8-16GB): ESM2-150M, ESM2-650M, ProtBert series
- High-end GPU (24GB+): ESM2-3B, ProtT5-XL, Ankh-large
- Multiple GPUs: ESM2-15B, ProtT5-XXL
Based on Task Type:
- Sequence classification: ESM2, ProtBert
- Structure prediction: ESM2, Ankh
- Generation tasks: ProtT5
- Antibody design: IgBert, IgT5
- Lightweight deployment: ProSST, PETA-base
Based on Training Data:
- General protein tasks: ESM2, ProtBert
- Structure-aware tasks: Ankh
- Antibody-specific: IgBert, IgT5
- Custom tokenization needs: PETA series

🔍 All models are available through the Hugging Face Hub and can be easily loaded using their templates.

🔬 Supported Training Approaches

Supported Training Approaches

Approach	Full-tuning	Freeze-tuning	SES-Adapter	AdaLoRA	QLoRA	LoRA	DoRA	IA3
Supervised Fine-Tuning	✅	✅	✅	✅	✅	✅	✅	✅

📚 Supported Datasets

Pre-training datasets

dataset	data level	link
CATH_V43_S40	structures	CATH_V43_S40
AGO_family	structures	AGO_family

Zero-shot datasets

dataset	task	link
VenusMutHub	mutation effects prediction	VenusMutHub
ProteinGym	mutation effects prediction	ProteinGym

Supervised fine-tuning datasets (amino acid sequences/ foldseek sequences/ ss8 sequences)

dataset	task	data level	problem type	link
DeepLocBinary	localization	protein-wise	single_label_classification	DeepLocBinary_AlphaFold2, DeepLocBinary_ESMFold
DeepLocMulti	localization	protein-wise	multi_label_classification	DeepLocMulti_AlphaFold2, DeepLocMulti_ESMFold
DeepLoc2Multi	localization	protein-wise	single_label_classification	DeepLoc2Multi_AlphaFold2, DeepLoc2Multi_ESMFold
DeepSol	solubility	protein-wise	single_label_classification	DeepSol_ESMFold
DeepSoluE	solubility	protein-wise	single_label_classification	DeepSoluE_ESMFold
ProtSolM	solubility	protein-wise	single_label_classification	ProtSolM_ESMFold
eSOL	solubility	protein-wise	regression	eSOL_AlphaFold2, eSOL_ESMFold
DeepET_Topt	optimum temperature	protein-wise	regression	DeepET_Topt_AlphaFold2, DeepET_Topt_ESMFold
EC	function	protein-wise	multi_label_classification	EC_AlphaFold2, EC_ESMFold
GO_BP	function	protein-wise	multi_label_classification	GO_BP_AlphaFold2, GO_BP_ESMFold
GO_CC	function	protein-wise	multi_label_classification	GO_CC_AlphaFold2, GO_CC_ESMFold
GO_MF	function	protein-wise	multi_label_classification	GO_MF_AlphaFold2, GO_MF_ESMFold
MetalIonBinding	binding	protein-wise	single_label_classification	MetalIonBinding_AlphaFold2, MetalIonBinding_ESMFold
Thermostability	stability	protein-wise	regression	Thermostability_AlphaFold2, Thermostability_ESMFold

✨ Only structural sequences are different for the same dataset, for example, DeepLocBinary_ESMFold and DeepLocBinary_AlphaFold2 share the same amino acid sequences, this means if you only want to use the aa_seqs, both are ok!

Supervised fine-tuning datasets (amino acid sequences)

dataset	task	data level	problem type	link
Demo_Solubility	solubility	protein-wise	single_label_classification	Demo_Solubility
DeepLocBinary	localization	protein-wise	single_label_classification	DeepLocBinary
DeepLocMulti	localization	protein-wise	multi_label_classification	DeepLocMulti
DeepLoc2Multi	localization	protein-wise	single_label_classification	DeepLoc2Multi
DeepSol	solubility	protein-wise	single_label_classification	DeepSol
DeepSoluE	solubility	protein-wise	single_label_classification	DeepSoluE
ProtSolM	solubility	protein-wise	single_label_classification	ProtSolM
eSOL	solubility	protein-wise	regression	eSOL
DeepET_Topt	optimum temperature	protein-wise	regression	DeepET_Topt
EC	function	protein-wise	multi_label_classification	EC
GO_BP	function	protein-wise	multi_label_classification	GO_BP
GO_CC	function	protein-wise	multi_label_classification	GO_CC
GO_MF	function	protein-wise	multi_label_classification	GO_MF
MetalIonBinding	binding	protein-wise	single_label_classification	MetalIonBinding
Thermostability	stability	protein-wise	regression	Thermostability
PaCRISPR	CRISPR	protein-wise	single_label_classification	PaCRISPR
PETA_CHS_Sol	solubility	protein-wise	single_label_classification	PETA_CHS_Sol
PETA_LGK_Sol	solubility	protein-wise	single_label_classification	PETA_LGK_Sol
PETA_TEM_Sol	solubility	protein-wise	single_label_classification	PETA_TEM_Sol
SortingSignal	sorting signal	protein-wise	single_label_classification	SortingSignal
FLIP_AAV	mutation	protein-site	regression
FLIP_AAV_one-vs-rest	mutation	protein-site	single_label_classification	FLIP_AAV_one-vs-rest
FLIP_AAV_two-vs-rest	mutation	protein-site	single_label_classification	FLIP_AAV_two-vs-rest
FLIP_AAV_mut-des	mutation	protein-site	single_label_classification	FLIP_AAV_mut-des
FLIP_AAV_des-mut	mutation	protein-site	single_label_classification	FLIP_AAV_des-mut
FLIP_AAV_seven-vs-rest	mutation	protein-site	single_label_classification	FLIP_AAV_seven-vs-rest
FLIP_AAV_low-vs-high	mutation	protein-site	single_label_classification	FLIP_AAV_low-vs-high
FLIP_AAV_sampled	mutation	protein-site	single_label_classification	FLIP_AAV_sampled
FLIP_GB1	mutation	protein-site	regression
FLIP_GB1_one-vs-rest	mutation	protein-site	single_label_classification	FLIP_GB1_one-vs-rest
FLIP_GB1_two-vs-rest	mutation	protein-site	single_label_classification	FLIP_GB1_two-vs-rest
FLIP_GB1_three-vs-rest	mutation	protein-site	single_label_classification	FLIP_GB1_three-vs-rest
FLIP_GB1_low-vs-high	mutation	protein-site	single_label_classification	FLIP_GB1_low-vs-high
FLIP_GB1_sampled	mutation	protein-site	single_label_classification	FLIP_GB1_sampled
TAPE_Fluorescence	fluorescence	protein-site	regression	TAPE_Fluorescence
TAPE_Stability	stability	protein-site	regression	TAPE_Stability

📈 Supported Metrics

Supported Metrics

Name	Torchmetrics	Problem Type
accuracy	Accuracy	single_label_classification/ multi_label_classification
recall	Recall	single_label_classification/ multi_label_classification
precision	Precision	single_label_classification/ multi_label_classification
f1	F1Score	single_label_classification/ multi_label_classification
mcc	MatthewsCorrCoef	single_label_classification/ multi_label_classification
auc	AUROC	single_label_classification/ multi_label_classification
f1_max	F1ScoreMax	multi_label_classification
spearman_corr	SpearmanCorrCoef	regression
mse	MeanSquaredError	regression

🤖 Agent-0.1: Your AI Assistant

Agent-0.1 is an intelligent AI assistant integrated into the VenusFactory platform, designed to answer questions and provide in-depth analysis on protein engineering and bioinformatics. It acts as a specialized expert, helping both biologists and AI researchers streamline their research workflow.

Key Features:

Zero-shot Prediction: Directly utilize cutting-edge sequence-based (e.g., ESM-2, ESM-1v, ESM-1b) and structure-based models (e.g., SaProt, ProtSSN, ESM-IF1, MIF-ST, ProSST) to perform zero-shot mutation prediction.
Protein Function: Accurately predict various protein functions, including solubility, localization, metal ion binding, stability, sorting signal, and optimum temperature.
Clear Insights: Always provides clear, actionable insights in response to your queries.

💡 Note: This feature requires an API key to access and is currently in Beta.

⚡ Quick Tools: Your Go-to for Rapid Predictions

Quick Tools is designed for users who need fast, efficient, and straightforward analysis without extensive configuration. It provides a no-code entry point to two key prediction tasks.

Directed Evolution: AI-Powered Mutation Prediction This tool allows for the rapid scoring and analysis of protein mutations. Simply upload a PDB file or paste the PDB content, and the platform will provide insights into the effects of single or multiple mutations on the protein.
Protein Function Leveraging pre-trained models, this module predicts various protein functions from a given amino acid sequence. You can upload a FASTA file or paste the sequence directly to predict properties such as solubility, localization, and more.

🧪 Advanced Tools: For In-depth Protein Analysis

Advanced Tools is built for researchers who require more granular control and deeper analysis. It offers powerful zero-shot prediction capabilities by allowing you to choose between two distinct model types.

Sequence-based Model This submodule focuses on high-throughput mutation effect scoring using powerful sequence-only models like ESM-2. You can upload a FASTA file or paste a protein sequence to perform large-scale predictions and score mutations.
Structure-based Model For tasks that require a deep understanding of protein 3D geometry, this tool utilizes structure-aware models like ESM-IF1. By uploading a PDB file or pasting its content, you can perform sophisticated zero-shot predictions that take the protein's spatial context into account.

✈️ Requirements

Hardware Requirements

Recommended: NVIDIA RTX 3090 (24GB) or better
Actual requirements depend on your chosen protein language model

Software Requirements

Anaconda3 or Miniconda3
Python 3.12

📦 Installation Guide

Git start with macOS

To achieve the best performance and experience, we recommend using Mac devices with M-series chips (such as M1, M2, M3, etc.).

1️⃣ Clone the repository

First, get the VenusFactory code:

git clone https://github.com/AI4Protein/VenusFactory.git
cd VenusFactory

2️⃣ Create a Conda environment

Ensure you have Anaconda or Miniconda installed. Then, create a new environment named venus with Python 3.12:

conda create -n venus python=3.12
conda activate venus

3️⃣ Install Pytorch and PyG dependencies

# Install PyTorch
pip install --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu

# Install PyG dependencies
pip install torch_scatter torch-sparse torch-geometric -f https://data.pyg.org/whl/torch-2.8.0+cpu.html

4️⃣ Install remaining dependencies

Install the remaining dependencies using requirements_for_macOS.txt:

pip install -r requirements_for_macOS.txt

Git start with Windows or Linux on CUDA 12.8

We recommend using CUDA 12.8

1️⃣ Clone the repository

First, get the VenusFactory code:

git clone https://github.com/AI4Protein/VenusFactory.git
cd VenusFactory

2️⃣ Create a Conda environment

Ensure you have Anaconda or Miniconda installed. Then, create a new environment named venus with Python 3.12:

conda create -n venus python=3.12
conda activate venus

3️⃣ Install Pytorch and PyG dependencies

# Install PyTorch
pip install torch==2.8.0 torchvision --index-url https://download.pytorch.org/whl/cu128

# Install PyG dependencies
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.8.0+cu128.html

4️⃣ Install remaining dependencies

Install the remaining dependencies using requirements.txt:

pip install -r requirements.txt

Git start with Windows or Linux on CUDA 11.x

We recommend using CUDA 11.8 or later versions, as they support higher versions of PyTorch, providing a better experience.

1️⃣ Clone the repository

First, get the VenusFactory code:

git clone https://github.com/AI4Protein/VenusFactory.git
cd VenusFactory

2️⃣ Create a Conda environment

Ensure you have Anaconda or Miniconda installed. Then, create a new environment named venus with Python 3.12:

conda create -n venus python=3.12
conda activate venus

3️⃣ Install Pytorch and PyG dependencies

# Install PyTorch
pip install torch==2.7.0 --index-url https://download.pytorch.org/whl/cu118

# Install PyG dependencies
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.7.0+cu118.html

4️⃣ Install remaining dependencies

Install the remaining dependencies using requirements.txt:

pip install -r requirements.txt

Git start with Windows or Linux on CPU

1️⃣ Clone the repository

First, get the VenusFactory code:

git clone https://github.com/AI4Protein/VenusFactory.git
cd VenusFactory

2️⃣ Create a Conda environment

Ensure you have Anaconda or Miniconda installed. Then, create a new environment named venus with Python 3.12:

conda create -n venus python=3.12
conda activate venus

3️⃣ Install Pytorch and PyG dependencies

# Install PyTorch
pip install torch torchvision --index-url https://download.pytorch.org/whl/cpu

# Install PyG dependencies
pip install torch_geometric
pip install pyg_lib torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.8.0+cpu.html

4️⃣ Install remaining dependencies

Install the remaining dependencies using requirements.txt:

pip install -r requirements.txt

🚀 Quick Start with Venus Web UI

Start Venus Web UI

Get started quickly with our intuitive graphical interface powered by Gradio:

python ./src/webui.py

This will launch the Venus Web UI where you can:

Configure and run fine-tuning experiments
Monitor training progress
Evaluate models
Visualize results

Using Each Tab

We provide a detailed guide to help you navigate through each tab of the Venus Web UI.

1. Training Tab: Train your own protein language model

Select a protein language model from the dropdown menu. Upload your dataset or select from available datasets and choose metrics appropriate for your problem type.

Choose a training method (Freeze, SES-Adapter, LoRA, QLoRA etc.) and configure training parameters (batch size, learning rate, etc.).

Click "Start Training" and monitor progress in real-time.

Click "Download CSV" to download the test metrics results.

2. Evaluation Tab: Evaluate your trained model within a benchmark

Load your trained model by specifying the model path. Select the same protein language model and model configs used during training. Select a test dataset and configure batch size. Choose evaluation metrics appropriate for your problem type. Finally, click "Start Evaluation" to view performance metrics.

3. Prediction Tab: Use your trained model to predict samples

Load your trained model by specifying the model path. Select the same protein language model and model configs used during training.

For single sequence: Enter a protein sequence in the text box.

For batch prediction: Upload a CSV file with sequences.

Click "Predict" to generate and view results.

4. Download Tab: Collect data from different sources with high efficiency

AlphaFold2 Structures: Enter UniProt IDs to download protein structures
UniProt: Search for protein information using keywords or IDs
InterPro: Retrieve protein family and domain information
RCSB PDB: Download experimental protein structures

5. Manual Tab: Detailed documentation and guides

Select a language (English/Chinese).

Navigate through the documentation using the table of contents and find step-by-step guides.

🧬 Code-line Usage

For users who prefer command-line interface, we provide comprehensive script solutions for different scenarios.

Training Methods: Various fine-tuning approaches for different needs

Full Model Fine-tuning

# Freeze-tuning: Train only specific layers while freezing others
bash ./script/train/train_plm_vanilla.sh

Parameter-Efficient Fine-tuning (PEFT)

# SES-Adapter: Selective and Efficient adapter fine-tuning
bash ./script/train/train_plm_ses-adapter.sh

# AdaLoRA: Adaptive Low-Rank Adaptation
bash ./script/train/train_plm_adalora.sh

# QLoRA: Quantized Low-Rank Adaptation
bash ./script/train/train_plm_qlora.sh

# LoRA: Low-Rank Adaptation
bash ./script/train/train_plm_lora.sh

# DoRA: Double Low-Rank Adaptation
bash ./script/train/train_plm_dora.sh

# IA3: Infused Adapter by Inhibiting and Amplifying Inner Activations
bash ./script/train/train_plm_ia3.sh

Training Method Comparison

Method	Memory Usage	Training Speed	Performance
Freeze	Low	Fast	Good
SES-Adapter	Medium	Medium	Better
AdaLoRA	Low	Medium	Better
QLoRA	Very Low	Slower	Good
LoRA	Low	Fast	Good
DoRA	Low	Medium	Better
IA3	Very Low	Fast	Good

Model Evaluation: Comprehensive evaluation tools

Basic Evaluation

# Evaluate model performance on test sets
bash ./script/eval/eval.sh

Available Metrics

Classification: accuracy, precision, recall, F1, MCC, AUC
Regression: MSE, Spearman correlation
Multi-label: F1-max

Visualization Tools

Training curves
Confusion matrices
ROC curves
Performance comparison plots

Structure Sequence Tools: Process protein structure information

ESM Structure Sequence

# Generate structure sequences using ESM-3
bash ./script/get_get_structure_seq/get_esm3_structure_seq.sh

Secondary Structure

# Predict protein secondary structure
bash ./script/get_get_structure_seq/get_secondary_structure_seq.sh

Features:

Support for multiple sequence formats
Batch processing capability
Integration with popular structure prediction tools

Data Collection Tools: Multi-source protein data acquisition

Format Conversion

# Convert CIF format to PDB
bash ./script/tools/file/convert/maxit.sh

Metadata Collection

# Download metadata from RCSB PDB
bash ./script/tools/search/database/rcsb/download_rcsb_meta.sh

Sequence Data

# Download protein sequences from UniProt
bash ./script/tools/search/database/uniprot/download_uniprot_seq.sh

Structure Data

# Download from AlphaFold2 Database
bash ./script/tools/search/database/alphafold/download_alphafold_structure.sh

# Download from RCSB PDB
bash ./script/tools/search/database/rcsb/download_rcsb_structure.sh

Features:

Automated batch downloading
Resume interrupted downloads
Data integrity verification
Multiple source support
Customizable search criteria

Supported Databases

Database	Data Type	Access Method	Rate Limit
AlphaFold2	Structures	REST API	Yes
RCSB PDB	Structures	FTP/HTTP	No
UniProt	Sequences	REST API	Yes
InterPro	Domains	REST API	Yes

Usage Examples: Common scenarios and solutions

Training Example

# Train a protein solubility predictor using ESM2
bash ./script/train/train_plm_lora.sh \
    --model "facebook/esm2_t33_650M_UR50D" \
    --dataset "DeepSol" \
    --batch_size 32 \
    --learning_rate 1e-4

Evaluation Example

# Evaluate the trained model
bash ./script/eval/eval.sh \
    --model_path "path/to/your/model" \
    --test_dataset "DeepSol_test"

Data Collection Example

# Download structures for a list of UniProt IDs
bash ./script/tools/search/database/alphafold/download_alphafold_structure.sh

💡 All scripts support additional command-line arguments for customization. Use --help with any script to see available options.

🙌 Citation

Please cite our work if you have used our code or data.

@inproceedings{tan2025venusfactory,
  title={VenusFactory: An Integrated System for Protein Engineering with Data Retrieval and Language Model Fine-Tuning},
  author={Tan, Yang and Liu, Chen and Gao, Jingyuan and Banghao, Wu and Li, Mingchen and Wang, Ruilin and Zhang, Lingrong and Yu, Huiqun and Fan, Guisheng and Hong, Liang and others},
  booktitle={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations)},
  pages={230--241},
  year={2025}
}

🎊 Acknowledgement

Thanks the support of Liang's Lab.

Name		Name	Last commit message	Last commit date
Latest commit History 704 Commits
.install		.install
ckpt		ckpt
data		data
docker		docker
docs/manual		docs/manual
example		example
img		img
script/tools		script/tools
src		src
.dockerignore		.dockerignore
.env.example		.env.example
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
README_CN.md		README_CN.md
Scripts_notebook.ipynb		Scripts_notebook.ipynb
docker-compose.yaml		docker-compose.yaml
environment.yaml		environment.yaml
frpc_linux_amd64_v0.3		frpc_linux_amd64_v0.3
install.py		install.py
install_config.json		install_config.json
pyproject.toml		pyproject.toml
requirements.txt		requirements.txt
requirements_for_macOS.txt		requirements_for_macOS.txt

Folders and files

Latest commit

History

Repository files navigation

✏️ Table of Contents

📑 Features

🧬 Supported Models (Zero-shot prediction)

Sequence-structure models

Structure-only models

Sequence-only models

🤖 Supported Models (Training for supervised tasks)

Pre-training Protein Language Models

BPE Tokenization Series

Unigram Tokenization Series

Model Selection Guide

🔬 Supported Training Approaches

📚 Supported Datasets

📈 Supported Metrics

🤖 Agent-0.1: Your AI Assistant

Key Features:

⚡ Quick Tools: Your Go-to for Rapid Predictions

🧪 Advanced Tools: For In-depth Protein Analysis

✈️ Requirements

Hardware Requirements

Software Requirements

📦 Installation Guide

To achieve the best performance and experience, we recommend using ​Mac devices with M-series chips (such as M1, M2, M3, etc.).

1️⃣ Clone the repository

2️⃣ Create a Conda environment

3️⃣ Install Pytorch and PyG dependencies

4️⃣ Install remaining dependencies

We recommend using CUDA 12.8

1️⃣ Clone the repository

2️⃣ Create a Conda environment

3️⃣ Install Pytorch and PyG dependencies

4️⃣ Install remaining dependencies

We recommend using CUDA 11.8 or later versions, as they support higher versions of PyTorch, providing a better experience.

1️⃣ Clone the repository

2️⃣ Create a Conda environment

3️⃣ Install Pytorch and PyG dependencies

4️⃣ Install remaining dependencies

1️⃣ Clone the repository

2️⃣ Create a Conda environment

3️⃣ Install Pytorch and PyG dependencies

4️⃣ Install remaining dependencies

🚀 Quick Start with Venus Web UI

Start Venus Web UI

Using Each Tab

🧬 Code-line Usage

Full Model Fine-tuning

Parameter-Efficient Fine-tuning (PEFT)

Training Method Comparison

Basic Evaluation

Available Metrics

Visualization Tools

ESM Structure Sequence

Secondary Structure

Format Conversion

Metadata Collection

Sequence Data

Structure Data

Supported Databases

Training Example

Evaluation Example

Data Collection Example

🙌 Citation

🎊 Acknowledgement

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Uh oh!

Contributors

To achieve the best performance and experience, we recommend using Mac devices with M-series chips (such as M1, M2, M3, etc.).

Packages