ATOM Modeling PipeLine (AMPL) for Drug Discovery: A Hands-On Tutorial

Do you want to know how to use Machine Learning (ML) for accelerating drug discovery? These recordings and support materials developed by bioinformaticians introduce AMPL and AMPL’s capabilities for creating ML-ready datasets. The ATOM Modeling PipeLine (AMPL), is an open-source Conda-based software that automates key drug discovery steps. AMPL is designed to take molecular binding data (for example, IC50, ki, and so on) and carry out the ML steps with minimal user intervention.

The workshop included two sections: a short presentation and a hands-on tutorial.

Section 1:

A 20-minute presentation covers the following topics:

• Introduction to small-molecule binding and the database sources
• Issues associated with data ingestion and curation
• Exploratory data analysis of the ingested and curated datasets
• Use of different featurization methods like molecular fingerprints or properties (Molecular Weight, number of hydrogen-bond acceptors, and so on)
• Creation of ML-ready datasets

Section 2:

A 35-minute AMPL code demonstration is followed by a 5-minute Q&A. The presenter shares a Python Jupyter notebook that covered the following ML steps: data ingestion/curation, featurization, and visualization to create ML-ready datasets.

Here are the key sections of the notebook:

• Highlights of AMPL functions that are designed to address the common issues encountered during the data ingestion and curation of drug discovery or small-molecule-focused projects
• Introduction of the extensible AMPL featurizer module and a demonstration on how simple keyword choices can lead to the computation of a range of different feature sets
• Exploratory Data Analysis and visualization code templates that can be adopted for other drug discovery projects with very little modification

To learn more about the software, visit the AMPL GitHub repository (https://github.com/ATOMconsortium/AMPL).