空间站广场

论文

Notebooks

比赛

课程

Apps

我的主页

我的Notebooks

我的论文库

我的足迹

我的工作空间

任务

节点

文件

数据集

镜像

项目

数据库

公开

DiffDock Notebook

PyTorch

我是地球人

发布于 2023-07-19

数据集

AI4SCUP-CNS-BBB(v1)

©️ Copyright 2023 @ Authors
作者： 杨舒文 📨 李子尧 📨
日期：2023-07-20
共享协议：本作品采用知识共享署名-非商业性使用-相同方式共享 4.0 国际许可协议进行许可。
快速开始：点击上方的 开始连接 按钮，选择 diffdock:2023-07-20-4镜像 和任意配置机型即可开始。

代码

文本

DiffDock Notebook

代码

文本

1. Introduction

Paper on arXiv

GitHub

Bohrium implementation of DiffDock, state-of-the-art method for molecular docking, by Gabriele Corso*, Hannes Stark*, Bowen Jing*, Regina Barzilay and Tommi Jaakkola.

Overview

You might also be interested in this awesome interactive online tool by Simon Duerr on Hugging Face for running DiffDock and visualising the predicted structures on your browser:

代码

文本

2. Inference

代码

文本

[ ]

import os

os.chdir("/root")

os.getcwd()

代码

文本

2.1 PDB + SMILES Input

We provide two easy ways to input ONE complex by configuring:

complex_name (str): name that the complex will be saved with
one of
- protein_path (str): path to the protein file
- protein_sequence (str): sequence of the protein for ESMFold
ligand_description (str): either a SMILES string or the path to a molecule file that rdkit can read

代码

文本

[ ]

# with sequences

complex_name = "1a0q"

protein_path = None

protein_sequence = "DIELTQSPSSLSASLGGKVTITCKASQDIKKYIGWYQHKPGKQPRLLIHYTSTLLPGIPSRFRGSGSGRDYSFSISNLEPEDIATYYCLQYYNLRTFGGGTKLEIKRADAAPTVSIFPPSSEQLTSGGASVVCFLNNFYSKDINVKWKIDGSERQNGVLNSWTDQDSKDSTYSMSSTLTLTKDEYERHNSYTCEATHKTSTSPIVKSFNRNE:EVQLQESDAELVKPGASVKISCKASGYTFTDHVIHWVKQKPEQGLEWIGYISPGNGDIKYNEKFKGKATLTADKSSSTAYMQLNSLTSEDSAVYLCKRGYYGRSNVDYWGQGTTLTVSSAKTTPPSVYPLAPGSAAQTNSMVTLGCLVKGYFPEPVTVTWNSGSLSSGVHTFPAVLQSDLYTLSSSVTVPSSTWPSETVTCNVAHPASSTKVDKKIE"

ligand_description = "COc(cc1)ccc1C#N"

代码

文本

[ ]

# with file path

complex_name = "1a0q"

protein_path = "/root/data/1a0q/1a0q_protein_processed.pdb"

protein_sequence = None

ligand_description = "/root/data/1a0q/1a0q_ligand.sdf"

代码

文本

If inference of multiple complexes is required, please input a path of protein_ligand_csv with following format:

complex_name	protein_path	ligand_description	protein_sequence
1	/root/data/1a0q/1a0q_protein_processed.pdb	data/1a0q/1a0q_ligand.sdf
2	/root/data/1a0q/1a0q_protein_processed.pdb	COc(cc1)ccc1C#N

代码

文本

[ ]

# with csv path

protein_ligand_csv_path = "/root/data/protein_ligand_example_csv.csv"

代码

文本

2.2 Output Configuration and Hyper-parameter

代码

文本

[ ]

# Directory where the outputs will be written to

out_dir = "/data/DiffDock-results/user_inference"

# Whether to save a pdb file with all the steps of the reverse diffusion

save_visualisation = False

# Number of samples to generate

samples_per_complex = 10

# Batch size

batch_size = 32

# If True, use no noise in the final step of the reverse diffusion

no_final_step_noise = False

# Number of denoising steps

inference_steps = 20

# Number of denoising steps that are actually performed, equal to inference_steps if None

actual_steps = None

# Model Info (Default from https://github.com/gcorso/DiffDock/tree/main/workdir)

model_dir = '/root/workdir/paper_score_model'

ckpt = 'best_ema_inference_epoch_model.pt'

confidence_model_dir = '/root/workdir/paper_confidence_model'

confidence_ckpt = 'best_model_epoch75.pt'

代码

文本

2.3 Protein-ligand Docking

代码

文本

[ ]

from diffdock import inference

import warnings

warnings.filterwarnings("ignore")

inference(

protein_ligand_csv=protein_ligand_csv_path,

complex_name=complex_name,

protein_path=protein_path,

protein_sequence=protein_sequence,

ligand_description=ligand_description,

out_dir=out_dir,

save_visualisation=save_visualisation,

samples_per_complex=samples_per_complex,

batch_size=batch_size,

no_final_step_noise=no_final_step_noise,

inference_steps=inference_steps,

actual_steps=actual_steps,

model_dir=model_dir,

ckpt=ckpt,

confidence_model_dir=confidence_model_dir,

confidence_ckpt=confidence_ckpt,

)

代码

文本

PyTorch

点个赞吧