Rdkit I: using rdkit to screen the structural characteristics of chemical small molecules

         Recently, I am working on the project of small molecule screening , It involves some processing methods , Later, I will summarize some problems and solutions one by one .

         First, a brief introduction RDkit,RDkit It is used to deal with small chemical molecules python Open source package , It was first developed by Novartis , The bottom is made up of C++ Compiling . stay github There is its source code and instructions on , The address is as follows ：

https://github.com/rdkit/rdkithttps://github.com/rdkit/rdkit        RDkit stay Anaconda or miniconda Installation in environment ：

conda install -yq -c rdkit rdkit

         Here's a brief introduction smiles code ,smiles Coding is essentially a kind of coding that uses strings to express the two-dimensional structure of small molecules , adopt rdkit package , Can be chemdraw in .mol The small molecule structural formula of the format , or sdf The spatial coordinates of the format are converted into smiles code , Thus, it is brought into the machine learning and deep learning models for learning .

Now let's get to the point ：

         In onemillion small chemical molecules , Screening , Just stay NO2<2 individual ,Cl < 3 individual ,Br < 2 individual ,F < 6 individual , Number of aromatic rings < 5 Molecules of .

To solve this problem , Two... Are needed RDkit package , Here is only the simplest usage , If you are interested, you can see the package introduction or source code ：

1. rdkit.Chem.Lipinski

Lipinski( Lippings rule ) It is a common constraint rule of small molecule drugs , stay rdkit Of Lipinski The package contains the calculation of various parameters , Just know the small molecules smiles code , You can analyze its ,HeavyAtomCount,NumAromaticRings…… And so on , Some of the more common ones are :NumAromaticRings,NumHAcceptors,NumHDonors,NumRotatableBonds These kinds of .

2.rdkit.Chem

rdkit.Chem The package contains functions for operating small molecule objects , Including atomic operations , Key operation , Ring operation , Pharmacophore search and other functions .

Our main atomic operations here , Atomic operations involve functions including ：

• Traverse the atom ：m.GetAtoms()
• Get atomic index ：GetIdx()
• Get atomic serial number ：GetAtomicNum()
• Get atomic symbol ：GetSymbol()
• Get the number of atomic connections （ suffer H Whether to hide the influence ）：GetDegree()
• Get the total number of atomic connections （ And H Whether to hide or not is irrelevant ）：GetTotalDegree()
• Get the atomic form charge ：GetFormalCharge()
• Get the atomic hybridization method ：GetHybridization()
• Get the atomic explicit valence ：GetExplicitValence()
• Get the implicit valence of atoms ：GetImplicitValence()
• Get the total valence of atoms ：GetTotalValence()

There is a Zhihu article to read , Write very well ：

RDKit| Molecular basis operation and pharmacophore search - You know List of articles Atomic operation key operation ring operation manual realization oxygen group pharmacophore search 1. Atomic operation in rdkit in , Every atom in a molecule is an object , You can get all kinds of information through the attributes and functions of atomic objects . Traverse the atom ：m.GetAtoms() Get atomic index ：GetIdx() a …https://zhuanlan.zhihu.com/p/143111689 Before you start coding , Let's introduce the format of input data , Because I found that for some contacts RDkit For the late old biochemist , It's better to write more carefully . The input data is molecular smiles code , as follows , All the molecules that need to be screened smiles Put the code in a csv or txt In the file , Column name is smiles：

  No verbosity , Code up ：

import pandas as pd
from rdkit import Chem
from rdkit.Chem import Lipinski

#  Read in the data 
df = pd.read_csv('smiles.csv')

#  Number of screening aromatic rings ：aromatic rings < 5
df['NumAromaticRings'] = df['smiles'].apply(lambda x:Lipinski.NumAromaticRings(Chem.MolFromSmiles(x)))
#  It can also be parallel , use parallel_apply, According to your own needs 
df = df.drop(df[df.NumAromaticRings >= 5].index)

#  Screening F,Cl,Br And other elements 
m = [Chem.MolFromSmiles(x) for x in df.smiles.tolist()]
#  Get a list of fluorine numbers 
num_F = []
for i in range(len(m)):
    F = [atom.GetSymbol() for atom in m[i].GetAtoms()].count('F')
    num_F.append(F)

#  Get the list of chlorine content 
num_Cl = []
for i in range(len(m)):
    Cl = [atom.GetSymbol() for atom in m[i].GetAtoms()].count('Cl')
    num_Cl.append(Cl)

#  Get a list of Bromine Numbers 
num_Br = []
for i in range(len(m)):
    Br = [atom.GetSymbol() for atom in m[i].GetAtoms()].count('Br')
    num_Br.append(Br)

# F,Br,CL Quantity can be filtered 
print(f'max_F:{max(num_F)};max_Cl:{max(num_Cl)};max_Br:{max(num_Br)}')

#  Condition screening 
transform = {'num_F':num_F,'num_Cl':num_Cl,'num_Br':num_Br}
atom_nums = pd.DataFrame(transform)
df = pd.concat([df,atom_nums],axis=1)
print(df.info())
df = df.drop(df[df.num_F >= 6].index)
df = df.drop(df[df.num_Cl >= 3].index)
df = df.drop(df[df.num_Br >= 2].index)

#  Screening NO2 Groups , Here is based on the character string search 
dff = df[df['smiles'].str.contains(pat='[N+](=O)[O-]',regex=False)]
#  Get to include NO2 Of the group smiles code dataframe form 
smiles = dff.smiles.tolist()
NO2 = [xcount('[N+](=O)[O-]') for x in n]    
transform2 = {'smiles':smiles,'num_NO2':num_NO2}
FG_nums = pd.DataFrame(transform2)
#  With smiles by inner And df Difference set 
df2 = df.append(FG_nums)
df2 = df2.drop_duplicates(subset=['smiles'],keep=False)

#  Save the final result 
df2.to_csv('result.csv',header = True, index = False)

One article a week , Next week's notice RDkit Dealing with clustering