chemap_video_l.mp4
ChemApp is a Streamlit-based application designed to support virtual screening and early-stage drug discovery workflows.
The application integrates machine learning, automated machine learning (AutoML), molecular fingerprinting, de novo molecular generation, drug-likeness assessment, molecular docking, and interactive visualization into a single computational platform.
ChemApp is primarily designed to run in a Linux environment and can also be used under Windows Subsystem for Linux (WSL). For convenience, the application can be launched from Windows using a .bat script and a desktop shortcut.
Scan QR code to launch app:
ChemApp provides an integrated workflow for computational screening and prioritization of potentially bioactive molecules.
The application combines ligand-based machine learning prediction with drug-likeness assessment and structure-based molecular docking.
The main workflow is:
Importantly, molecular descriptors used for drug-likeness assessment are not used as input features for the machine learning model. The ML model uses molecular fingerprints as its representation.
ChemApp allows users to create a machine learning classification model for molecular bioactivity prediction.
The model development workflow can retrieve experimental bioactivity data directly from ChEMBL using the ChEMBL API.
The user can provide a ChEMBL ID, after which the application retrieves the corresponding dataset and preprocesses it for machine learning.
ChemApp supports datasets based on activity measurements such as:
- Ki
- IC50
The retrieved data is subsequently processed and converted into molecular representations suitable for machine learning.
ChemApp uses AutoML to automate the process of model development.
Once a classification model is available, ChemApp provides several ways to obtain molecules for prediction.
A user can enter an individual molecule using its SMILES representation.
Example:
Cc1n2[C@@H](COc3cccc(c23)c1C(=O)c4cccc5ccccc45)CN6CCOCC6
The molecule is then processed and passed to the trained ML model.
ChemApp can also import a collection of molecules from a .csv file.
For example:
smiles,id
Cc1n2[C@@H](COc3cccc(c23)c1C(=O)c4cccc5ccccc45)CN6CCOCC6,WIN 55
CC(C)(C)[C@@H](CO)NC(=O)c1nn(c2c[n+]([O-])ccn2)c3[C@H]4C[C@H]4Cc13,Olorinab
CC(C)CN1C(=O)CN(Cc2ccc(cc2)c3ccc(F)c(CNC4CCOCC4)n3)C1=O,LEI102
Cc1c(nn(CCCCCO)c1c2ccccc2)C(=O)NC34CC5CC(CC(C5)C3)C4,AM10257This allows multiple compounds to be screened in a single workflow.
ChemApp can generate new molecular structures using ChemBERTaLM – Hugging Face Model.
The generated molecules can then be passed through the same screening pipeline as externally supplied compounds.
After molecules have been imported or generated, ChemApp can use the previously trained classification model to predict their bioactivity.
The ML prediction uses molecular fingerprints, consistent with the representation used during model training.
The predicted bioactivity can be used as an initial filtering or prioritization step before further computational analysis.
After bioactivity prediction, ChemApp calculates molecular descriptors used to assess drug-likeness.
These descriptors are not used as features by the ML model. They represent a separate stage of the workflow intended to characterize the physicochemical properties of candidate molecules.
ChemApp integrates molecular docking using AutoDock Vina.
Docking is performed for molecules after the preceding screening steps.
The purpose of this stage is to investigate potential interactions between candidate ligands and a selected protein target.
ChemApp provides interactive visualization of docking results using py3Dmol.
git clone https://github.com/Adam-maz/ChemApp.git
cd ChemAppCreate Conda environment:
conda env create -f chemapp_environment.ymlActivate it:
conda activate chemapp_envAfter activating the virtual environment, ChemApp can be started using:
streamlit run ChemApp.pyStreamlit will start a local web server.
The application is typically available at:
http://localhost:8501
The exact address and port are displayed by Streamlit when the application starts.
For convenience, ChemApp can be launched from Windows using a .bat file.
The batch script can start the application inside WSL, making it possible to launch ChemApp using a Windows desktop shortcut.
A simplified example is:
@echo off
wsl bash -c "cd /path_to_ChemApp && source .run_chemapp.sh"
pauseThe path should be adjusted to the actual location of the ChemApp project.
The .bat file can then be linked to a Windows desktop shortcut.
run_chemapp.sh script and .ico file are also provided
ChemApp is intended as a computational research and educational tool for virtual screening and early-stage drug discovery.
The results generated by the application are subject to the limitations of the underlying computational methods.
Potential limitations include:
- Dependence on the quality and representativeness of ChEMBL data
- Experimental noise in bioactivity measurements
- Potential bias in training datasets
- Applicability-domain limitations of ML models
- Limitations of molecular fingerprints
- Uncertainty associated with ML predictions
- Simplifications introduced by drug-likeness descriptors
- Approximate nature of molecular docking
- Limitations of docking scoring functions
- Limited representation of protein flexibility
- Possible generation of chemically undesirable structures by generative models
Therefore, ChemApp should be considered as a tool for molecular screening, candidate prioritization, and hypothesis generation rather than a replacement for experimental validation.
Experimental studies are required to confirm predicted biological activity and molecular binding.
ChemApp can be used for:
- Virtual screening
- Early-stage drug discovery
- Computational chemistry
- Cheminformatics
- Bioactivity prediction
- Machine learning model development
- Automated machine learning
- Molecular fingerprint-based classification
- De novo molecular generation
- Drug-likeness assessment
- Molecular docking
- Docking pose analysis
- Protein–ligand visualization
- Computational chemistry education