LEAD DISCOVERY

OpenEye Machine Learning and Molecule Explainer

Build machine learning models, assess model quality, visualize and understand the results – all in one environment.

Machine learning (ML) models are typically opaque to the user, producing predictions that are difficult, if not impossible to interpret.

Unlike other opaque approaches, the ML tools in Orion allow you to build models, assess quality, and understand your results in a chemically intuitive way.

In addition to building ML models, users also have access to pre-trained ML models for solubility and hERG toxicity.

OpenEye Machine Learning Molecule Explainer

With OpenEye’s Molecule Explainer, machine learning is no longer just a unknown-box system. Scientists can visualize areas of a molecule the algorithm considers important for the predicted property.

Features

Transparency. Explains predictions with molecular annotations
Confidence. Domain of applicability is provided with the predicted values
Performance. Extremely fast model building and computation
Customisability. Guided workflows and pre-trained models (solubility and hERG toxicity) are provided
Control. Highly parameterizable for novice use and expert control
Modularity. Integrate with other physics-based and cheminformatics methods and pipelines

OE ML Floe Results with Class Confidence

OpenEye’s machine learning results let you easily assess the quality of the prediction. For each prediction, a class confidence (low, medium, or high) is provided to help you determine trustworthiness of the predicted value.

OpenEye’s guided workflows (Orion® Floes) are designed to help you quickly build and assess your machine learning models. The built-in Molecule Explainer^(TM) feature provides you with an easy to understand visual explanation of the predicted physical properties for your small molecules.

OpenEye uses a fully connected and probabilistic neural network implemented in TensorFlow that makes machine learning faster and easier.

We have created a Molecule Explainer to provide visual insight into the predicted results at the fragment or atomic level. Molecule Explainer is an adaptation of the Local Interpretable Model agnostic Explanations (LIME [2]) technique, which maps machine learning predictions onto molecular structure.

OpenEye machine learning output report makes it easy for scientists to assess the quality of the predicted results. For each prediction, a class confidence (low, medium, or high) is assigned to the predicted value.

OpenEye's machine learning default parameters is convenient for novice user to build models, but can be highly parameterizable for expert control. In addition, visualization and statistics provided with each model can help you select best models for your dataset. With the 2023.1 release, it is also possible to update previously built models with new data.

OpenEye’s Orion, a web-based modeling platform, lets you seamlessly integrate both physics-based (classical and QM) models and data-driven (machine learning) models into your computational drug-design projects.

New features include:

Ability to add a custom feature vector for building machine learning models
Options to build classification or regression models.
Update previously built models with new data
Predict and explain hERG toxicity in an intuitive way

Learn More

OpenEye’s machine learning tools help you generate data-driven models, speed up your drug design process, and reduce cost by reducing failure rates.

Orion Small Molecule Discovery Machine Learning

And, you can also WATCH OpenEye’s miniWebinar recording from September 8, 2022 on machine learning with one of our domain expert scientists, Sayan Mandal, Ph.D.