Skip to content

Molecule Generate User Guide

1. Introduction to the algorithm

Generative chemistry is a very important for de-novo drug design, which can find new chemical scaffolds beyond existing screening libraries. The molecular generation algorithm of Tencent is to learn the structure information of small molecules related to protein targets in a known chemical space. Currently, we support 319 kinase and 52 GPCR targets to generative chemistry. During this process, our algorithm can sample molecules from the projection of the molecular space towards different targets and then generate novel molecules with activities. Or you can select a specific target and upload a reference compound. Our algorithm can also generate novel molecules by changing motifs of the reference compound, with maintaining bioactivity.

2. Start to Use

Two ways are supproted by our Generative chemistry model: only target or target with ONE reference compound.

2.1 Input target of interest only

Please select or type in one target that you are interested in from our target list. Then our model can generate one batch of molecules with their basic properties and predicted ADMET properties. You can select the compounds with your desired properties later.


2.2 target with one reference compound

By this way, you can refine the molecular generation process by inputting reference compound. Firstly select or input your target and then input the SMILES of reference compound or draw the structure of it. after the submission the task, our algorithm will generate a batch of novel molecules with basic and ADMET properties.


It usually will take 40 minutes for the generation molecules depends on the complexity of your reference compound. Please be patient.

3. Query the Result of Your Task

3.1 Query history

You can check at 'Recent Histroy' whether your submitted tastk is ready or not at the bottom of page and check the complete task.


3.2 Selection criteria for post-generation of molecules

You can select one or more properties of the generated molecules as you prefer to get your final compound list.

  • Check synthetic accessibility score, 2D/3D similarity as your reference compound, and predicted ADMET properties;


  • Clustering and filtering by properties;

    We prepared 8 common properties as below: logP(Partition coefficient)
    HBA(hydrogen bond acceptor),HBD(hydrogen bond donor)
    molecular weight
    TPSA (topological polar surface area)
    hERG(“negative” means the probability of hERG < 0.3; “positive” means the probability of hERG > 0.3)
    Blood_brain barrier permeability (“negative”means the probability < 0.5; “positive” means the probability >0.5)


  • All the information of generated molecules can be downloaded as CSV file for your later use.