★SMILES数据处理+评价指标所有的代码

首先看一下文件夹目录：

---

一、数据获取

（1）、QM9

获取地址为：https://github.com/nyu-dl/dl4chem-mgm

直接下载这个txt文件就可以了，然后直接把后缀改为csv 

（2）、ChEMBL

获取地址为：https://github.com/BenevolentAI/guacamol

 或者直接按照如图下的进行下载，然后把每个文件后缀改为csv

Download

You can download pre-built datasets here:

md5 05ad85d871958a05c02ab51a4fde8530 training
md5 e53db4bff7dc4784123ae6df72e3b1f0 validation
md5 677b757ccec4809febd83850b43e1616 test
md5 7d45bc95c33c10cb96ef5e78c38ac0b6 all

---

二、CSV文件转为images

运行的命令为：

python smiles2images.py --smiles_dataroot ./datasets_smiles/ChEMBL/guacamol_v1_all.csv --images_datafolder ./datasets_images/ChEMBL/ --smiles_first_line CCCC(=O)NNC(=O)Nc1ccccc1

代码为： 

import argparse
import os
import pandas as pd
from tqdm import tqdm
from rdkit import Chem
from rdkit.Chem.Draw import rdMolDraw2D


def main():
    parser = argparse.ArgumentParser(description='SMILES to images')
    # SMILES数据集所在的文件夹
    parser.add_argument('--smiles_dataroot', type=str, default="./datasets_smiles/ChEMBL/guacamol_v1_all.csv", help='data root')
    parser.add_argument('--images_datafolder', type=str, default="./datasets_images/ChEMBL/", help='data root')
    parser.add_argument('--smiles_first_line', type=str, default="CCCC(=O)NNC(=O)Nc1ccccc1", help='这是csv文件的第一行索引')
    args = parser.parse_args()
    # SMILES数据集
    raw_file_path = os.path.join(args.smiles_dataroot)
    # 输出图像文件夹
    opt_img = os.path.join(args.images_datafolder)

    if not os.path.exists(opt_img):
        os.makedirs(opt_img)

    df = pd.read_csv(raw_file_path)
    opt_smiles = df[args.smiles_first_line].values

    i = 1
    for opt_smiles in tqdm(opt_smiles, total=len(opt_smiles)):
        filename = "{}.png".format(i)
        i = i + 1

        opt_img_save_path = os.path.join(opt_img, filename)

        mol = Chem.MolFromSmiles(opt_smiles)
        drawer = rdMolDraw2D.MolDraw2DCairo(256, 256)
        opts = rdMolDraw2D.MolDrawOptions()
        # 设置杆的粗细
        opts.bondLineWidth = 3
        # 设置字母的大小
        opts.minFontSize = 15

        drawer.SetDrawOptions(opts)
        rdMolDraw2D.PrepareAndDrawMolecule(drawer, mol)
        drawer.FinishDrawing()
        drawer.WriteDrawingText(opt_img_save_path)


if __name__ == '__main__':
    main()

三、将生成的image识别为SMILES，然后将SMILES转换为image

这里直接看连接：生成的分子图像是否可以识别为SMILES，然后再将识别后的SMILES转换为图像？_马鹏森的博客-CSDN博客

四、评价三中生成转为的image（metrics）

运行的命令为：

python distribution_learning_file.py --generate_smiles_file image2smiles_validity.csv --train_smiles_file 100k_moleculars.csv --output_dir ./ --suite v2 --number_samples 10000

代码为：

import argparse
import os

from guacamol.assess_distribution_learning import _assess_distribution_learning
from guacamol.utils.helpers import setup_default_logger
from typing import List
from guacamol.distribution_matching_generator import DistributionMatchingGenerator


class AllGeneratorSmiles(DistributionMatchingGenerator):
    """
    Generator that samples SMILES strings from a predefined list.
    """

    def __init__(self, molecules: List[str]) -> None:
        """
        Args:
            molecules: list of molecules from which the samples will be drawn
        """
        self.molecules = molecules

    def generate(self, number_samples: int) -> List[str]:
        return list(self.molecules)


def main():
    setup_default_logger()

    parser = argparse.ArgumentParser(description='Molecule distribution learning benchmark for random smiles sampler', formatter_class=argparse.ArgumentDefaultsHelpFormatter)
    parser.add_argument('--generate_smiles_file', default='image2smiles_validity.csv', help='这个是生成的N个SMIMES文件')
    parser.add_argument('--train_smiles_file', default='100k_moleculars.csv', help='这个是训练集的SMIMES文件')
    parser.add_argument('--output_dir', default='./', help='Output directory')
    parser.add_argument('--suite', default='v2')
    parser.add_argument('--number_samples', default='10000', type=int, help='Number of samples to validity generate')
    args = parser.parse_args()

    if args.output_dir is None:
        args.output_dir = os.path.dirname(os.path.realpath(__file__))

    with open(args.generate_smiles_file, 'r') as smiles_file:
        smiles_list = [line.strip() for line in smiles_file.readlines()]
    generator = AllGeneratorSmiles(molecules=smiles_list)


    json_file_path = os.path.join(args.output_dir, 'distribution_learning_results.json')



    # number_samples是从生成的文件中随机取出多少个样本
    _assess_distribution_learning(model = generator,
                                 chembl_training_file = args.train_smiles_file,
                                 json_output_file = json_file_path,
                                 benchmark_version = args.suite,
                                 number_samples=args.number_samples)


if __name__ == '__main__':
    main()