rdkit分子指纹生成；相似性检索

1、分子指纹生成

常见生成的方法：

a、smiles2fp simles转分子指纹，支持morgan: Morgan fingerprint.
* maccs: MACCS keys.
* rdkit: RDKit topological fingerprint.三种的函数

from tqdm import tqdm
import numpy as npfrom rdkit import Chem
from rdkit.Chem import AllChem, Draw
from rdkit.Chem import MACCSkeysdef smiles2fp(smiles, fp_type):r"""Convert smile strings to fingerprint.Molecular fingerprints are a way to represent molecules as mathematical objects.Note:To use this trick, you should install [rdkit](https://www.rdkit.org/) at first:```bashpip install rdkit```Args:smiles (list of str): The smile strings to convert.fp_type (list of str): The types of generated fingerprint. Can be the following values:* `morgan`: Morgan fingerprint.* `maccs`: MACCS keys.* `rdkit`: RDKit topological fingerprint.Returns:(np.array): The generated fingerprint features."""fps = {fpt: [] for fpt in fp_type}for i in tqdm(range(len(smiles))):rdkit_mol = Chem.MolFromSmiles(smiles[i])for fpt in fp_type:if fpt == 'morgan':fp = AllChem.GetMorganFingerprintAsBitVect(rdkit_mol, 2)elif fpt == 'maccs':fp = MACCSkeys.GenMACCSKeys(rdkit_mol)elif fpt == 'rdkit':fp = Chem.RDKFingerprint(rdkit_mol)fps[fpt].append(fp)return [np.array(fps[fpt], dtype=np.int64) for fpt in fp_type]smiles_printvec = smiles2fp(smis, ['morgan'])

b、

from rdkit.Chem import AllChemd1 =d["ROMol"][0]
d2 =d["ROMol"][1]
##指纹创建
c1 = rdkit.Chem.RDKFingerprint(d1,maxPath=7,fpSize=512)
c2 = rdkit.Chem.RDKFingerprint(d2,maxPath=7,fpSize=512)
c1.ToBitString()### 另一种指纹生成方法
c5 = AllChem.GetMorganFingerprintAsBitVect(d1,5,nBits=512)
c6 = AllChem.GetMorganFingerprintAsBitVect(d2,5,nBits=512)
c6.ToBitString()from rdkit import DataStructs
##相似度计算
DataStructs.DiceSimilarity(c1,c2),DataStructs.DiceSimilarity(c5,c6)DataStructs.CosineSimilarity(c1,c2)

2、相似性检索

计算相似度方法：

a、

pubchempy包可以下载pubchem网站上分子，但速度慢
数据源：1.txt是找的一些pubchem 分子id值

import pubchempy as pcpwith open("1.txt","r") as f:alls = f.read().splitlines()all_mols=[]
for i in alls:c = pcp.Compound.from_cid(i)print(i)all_mols.append(c.isomeric_smiles)

下面用这些下载的分子做相似性计算：


from rdkit.Chem import AllChem, Draw from rdkit import DataStructs,Chemmols = [Chem.MolFromSmiles(x) for x in all_mols if x is not None]#计算分子库每个分子指纹
mols_fps = [(m, AllChem.GetMorganFingerprint(m, 2)) for m in mols]#读入要查询的单个分子，同样计算指纹
nicotine = Chem.MolFromSmiles('O=C(C)Oc1ccccc1C(=O)O')
nicotine_fingerprint = AllChem.GetMorganFingerprint(nicotine, 2)#计算相似度并排序，输出最相似的的前20个分子
mols_nicotinesim = [(m, DataStructs.TanimotoSimilarity(fp, nicotine_fingerprint))  for m, fp in mols_fps]sorted_mols_nicotinesim = sorted(mols_nicotinesim, key=lambda x: x[1], reverse=True)result = sorted_mols_nicotinesim[:20]# 画出
Draw.MolsToGridImage([m for m, sim in result], subImgSize=(200,200),molsPerRow=4, legends=['' for x in result])

单个要查询的输入分子：
查到的xiang相似前20：

b、另外方法，分子指纹onehot方式 Cosine similarity余弦相似度计算

这里读取1万个smi分子，文件下载可参考：https://github.com/zilliztech/MolSearch/edit/master/script/test_1w.smi

from rdkit import Chem
from rdkit.Chem import AllChem, Draw## 加载数据
mols2 = []
with open("1w.smi","r") as f:ss = f.readlines()for i in ss:mols2.append(i.strip().split()[0])

Draw.MolsToGridImage([ Chem.MolFromSmiles(mols2[i]) for i in range(10)],subImgSize=(300,300), molsPerRow=3)

这里Draw.MolsToGridImage函数第一个参数必须是mol格式的列表

另外每个图需要说明可以用legends 这参数，然后列表里的元素必须为字符串

from tqdm import tqdm
import numpy as np
from rdkit.Chem import MACCSkeysdef smiles2fp(smiles, fp_type):fps = {fpt: [] for fpt in fp_type}for i in tqdm(range(len(smiles))):for fpt in fp_type:if fpt == 'morgan':fp = AllChem.GetMorganFingerprintAsBitVect(smiles[i], 2,nBits=512)elif fpt == 'maccs':fp = MACCSkeys.GenMACCSKeys(rdkit_mol)elif fpt == 'rdkit':fp = Chem.RDKFingerprint(rdkit_mol)fps[fpt].append(fp)return [np.array(fps[fpt], dtype=np.int64) for fpt in fp_type]
## 计算1万个分子的指纹one hot
smiles_printvec = smiles2fp(mols2, ['morgan'])##带查询的阿司匹林分子指纹
aspirin= Chem.MolFromSmiles('CC(=O)OC1=CC=CC=C1C(O)=O')
aspirin_printvec = np.array(AllChem.GetMorganFingerprintAsBitVect(aspirin, 2,nBits=512)）##数据处理归一，相似度计算nnn = np.linalg.norm(smiles_printvec )* np.linalg.norm(aspirin_printvec )
iii = np.dot(smiles_printvec , aspirin_printvec .T)
sims = iii/nnn
sims## 取最相似的前10个，用squeeze进行降维度，因为argsort必须作用于一维[ii for ii in np.squeeze(sims).argsort()[::-1][:10]]

结果：阿司匹林

最相似的前10个
Draw.MolsToGridImage([ Chem.MolFromSmiles(mols2[i]) for i in [4786, 5047, 4231, 8489, 3417, 9829, 2685, 3063, 9201, 3326]],subImgSize=(300,300), molsPerRow=5)