The ATOM Modeling PipeLine (AMPL; https://github.com/ATOMconsortium/AMPL) is an open-source, modular, extensible software pipeline for building and sharing models to advance in silico drug discovery. To see the list of AMPL parameters, please check this link, https://github.com/ATOMconsortium/AMPL/blob/master/atomsci/ddm/docs/PARAMETERS.md
This page contains a collection of AMPL-COLAB tutorial notebooks.
+ Please note that if you have trouble opening up any of the following notebooks, please go to, https://nbviewer.jupyter.org/, and paste the notebook link to view the contents.
- Tutorial-00: Basic COLAB tutorial. For all the COLAB tutorials, click on the tutorial link, and then click on "Open in Colab" baner. You can open and run the notebook from the browser. If you want to save your edits to the notebook, you need to save a copy in your Google Drive. Usually, Google COLAB saves the notebook files under the "My Drive > Colab Notebooks" folder
The data that we gather for modeling is small-molecule/drug binding data. The following links will introduce some of the concepts and outcome measures related to this topic:
- https://en.wikipedia.org/wiki/IC50
- https://bpspubs.onlinelibrary.wiley.com/doi/pdfdirect/10.1111/j.1476-5381.2009.00604.x
For the tutorials, we will use the small-molecule binding data obtained from either one of the following resources, ChEMBL (https://www.ebi.ac.uk/chembl/), Drug Target Commons (DTC; https://drugtargetcommons.fimm.fi/) and Excape-DB (https://solr.ideaconsult.net/search/excape/).
Click here to learn about single-target focussed data.
-
Tutorial-01: (Time: ~ 3 minutes) This COLAB notebook will use AMPL for data cleaning, EDA and clustering on ExCAPE-DB (https://solr.ideaconsult.net/search/excape/) data for HTR3A protein
-
Tutorial-02: (Time: ~ 6 minutes) This COLAB notebook will use AMPL for data curation of HTR3A protein data from ExCAPE-DB (https://solr.ideaconsult.net/search/excape/) Data
- Tutorial-03: (Time: ~ 4 minutes) This COLAB notebook will use AMPL for Data cleaning, EDA and clustering of HTR3A protein data from Drug Target Commons (DTC)
- Tutorial-04: (Time: ~ 10 minutes) This COLAB notebook will use AMPL for Data curation of HTR3A protein data from Drug Target Commons (DTC)
- Tutorial-05: (Time: ~ 4 minutes) This COLAB notebook will use AMPL to upload datasets (small-molecule activity data from ChEMBL), clean, merge and do some basic Exploratory Data Analysis.
- Tutorial-06: (Time: ~ 4 minutes) This COLAB notebook with use AMPL to merge HTR3A binding data from two different data sources, DTC and ExCAPE-DB.
- Tutorial-07: (Time: ~ 4 minutes). The notebook uses HTR3A as the protein target. The notebook accomplishes the following tasks:
- Uses AMPL software
- Reads in data from three database sources: ChEMBL, Excape-DB and DTC
- Cleans, standardizes and analyzes the data
- Merges and harmonizes to create a dataset
- Tutorial-08: Exploratory Data Analysis-Regression
- Tutorial-09: Exploratory Data Analysis-Regression
- Tutorial-10: (Time: ~ 2 minutes): Simple supervised learning example. AMPL will read the public data (117 chemical compounds), curate, fit a Random Forest model to predict solubility and test the model. For additional information on the dataset, please check this publication,https://pubmed.ncbi.nlm.nih.gov/15154768/
- Tutorial-11: (Mode: AMPL_GPU; Time: ~ 18 minutes): This COLAB notebook will use AMPL for predicting binding affinities -pIC50 values- of ligands that could bind to human Sodium channel protein type 5 subunit alpha protein (Gene: SCN5A) using Graph Convolutional Network Model. ChEMBL database is the data source of binding affinities (pIC50)
3. Hyper-parameter Optimization (HPO), Uncertainty Quantification (UQ), and using metrics for analyzing model performance.
This notebook also explores AMPL functions for saving and loading prebuild AMPL models for analysis.
- Tutorial-12 Hyper-parameter Optimization (HPO) and Uncertainty Quantification (UQ).
- Tutorial-13 Notebook includes HPO Grid Search on three different modeling methods (Random Forest, NN and XGBoost).
- Tutorial-12 Notebook provides the framework for visualizing the results of HPO results and use them to identify best models.
- Tutorial-14: This notebook creates an AMPL (RF) model using BSEP dataset (reference: https://pubmed.ncbi.nlm.nih.gov/33502191/), and makes predictions (inference) on an external sample test dataset.
Workshop date, June 05, 2021: Protein Target-focussed Binding Data Curation, Exploratory Data Analysis and Featurization using AMPL. Please note that Google Chrome browser works best with the COLAB Jupyter notebooks
- Click here to access the presentation slides.
- Click here to open the tutorial Jupyter COLAB notebook. If for some reason, the notebook link doesnt work for you, please use this one, https://nbviewer.jupyter.org/github/ravichas/AMPL-Tutorial/blob/master/AMPL_FNL_Workshop_06052021.ipynb
- For use with GCP or other pre-installed AMPL environments: https://github.com/ravichas/AMPL-Tutorial/blob/master/GCP_AMPL_FNL_Workshop_06052021.ipynb
- DeepChem, https://deepchem.io/
- rdkit, https://www.rdkit.org/
- ChEMBL: https://www.ebi.ac.uk/chembl/
- PubChem: https://pubchem.ncbi.nlm.nih.gov/
- Drug Target Commons (DTC): https://drugtargetcommons.fimm.fi/
- ExCAPE-DB: https://solr.ideaconsult.net/search/excape/
- DrugBank: https://go.drugbank.com/
Most of the tutorial code chunks came from multple Jupyter notebooks generously shared by the ATOM team.
- Amanda Paulson
- Ben Madej
- Da Shi
- Hiran Ranganathan
- Jessica Mauvais
- Jonathan Allen
- Kevin Mcloughlin
- Sarangan Ravichandran
- Stewart He
- Ya Ju Fan
- Contributions from the following student programs:
- The Purdue Data Mine; https://datamine.purdue.edu/
- Butler University
- Columbia University