Legacy Release Notes
v0.12.1 February 2025
This package is built using OpenEye-orionplatform==6.2.0, OpenEye-toolkits==2024.2.0, OpenEye-Snowball==0.28.1, and OpenEye-floereport==6.2.0.
Feature Updates
There is now a baseline XGBoost model with every ML Build floe. The performance of these XGBoost models is reported in the Floe Report and output success model.
Several bug fixes have been made.
New Floe in this package
ML Predict: XGBoost for Small Molecules Floe
The XGBoost models built by default using ML Build floes can be used as input here. This floe can be used to predict the properties of small molecules using XGBoost from the ML Build floes. It will act as a classifier if the base model is a classifier trained on String and will act as a regressor if trained on Float. The floe autoselects the type of prediction based on the provided model. If the model was trained on a custom feature vector, please provide a field in the input molecule dataset containing the feature vector; otherwise, the floe will fail.
v0.12.0 July 2024
This package is built using
OpenEye-orionplatform==6.2.0,OpenEye-toolkits==2024.1.1,OpenEye-Snowball==0.28.0, andOpenEye-floereport==6.2.0.All Floes have new tutorial links.
Feature Updates
Scaffold splitting has been added to the Data Processing of Small Molecules for ML Model Building Floe.
The option to use a Keras Hyperparameter Tuner has been added to the ML Build: Regression Model with Tuner using Feature Input Floe.
A default Adjust Batch Size parameter has been added to the Advanced: Neural Network Hyperparameter Options for all ML Build Floes.
This release includes a new style for Floe Reports, bug fixes, and more stable Floe runs.
v0.11.1 July 2023
This package is built using
OpenEye-orionplatform==5.1.0,OpenEye-toolkits==2023.1.0, andOpenEye-Snowball==0.26.0.All Floes have new brief texts and Floe categories.
Feature Updates
Keras Hyperparameter Tuner.
This is an optimizer for fingerprint based model building (classification and regression).
Choose from Hyperband, RandomSearch, and Bayesian Optimization.
It is available in Floes: ML Build: Regression Model with Tuner using Fingerprints for Small Molecules and ML Build: Classification Model with Tuner using Fingerprints for Small Molecules.
This release includes a new style for Floe reports, bug fixes, and more stable Floe runs for larger data.
New Floes in this package
hERG Toxicity Prediction for Small Molecules using ML and Cheminfo Fingerprints.
This Floe predicts and explains hERG toxicity of molecules as either “High” or “Low.”
ML ReBuild: Transfer Learn ML Regression Model using Fingerprints for Small Molecules.
This Floe is for Transfer Learning on previously built Machine Learning Models in Orion.
v0.10.5 March 2023
Feature Updates
This is the first minor release of Model Building package with package version 0.10.4.
This package is built using
OpenEye-orionplatform==4.4.0,OpenEye-toolkits==2022.2.1, andOpenEye-Snowball==0.24.0..Added Overfit correlation and coefficients [refer to Optimization Strategy V] to train models without overfit.
Added Precision, Recall, Specificity, and Sensitivity parameters in floe report for Classification Model Building Floes. These fields are also available in record output.
v0.10.0 December 2022
Feature Updates
This is the second major release of Model Building package with package version 0.10.0.
This package is built using
OpenEye-orionplatform==4.5.4,OpenEye-toolkits==2022.2.1, andOpenEye-Snowball==0.24.0..All floes have a new brief parameter and floe categories.
New Floes in this package
ML Build: Classification Model with Tuner using Fingerprints for Small Molecules trains multiple Neural Network Classification models on physical properties of small molecules. Uses cheminformatics and machine learning to train said models.
ML Predict: Classification using Fingerprints for Small Molecules predicts discreet classes from a pretrained neural network model. Needs molecule database and neural network model dataset as input (trained using the previous floe).
ML Build: Regression Model with Tuner using Feature Input trains multiple Neural Network Regression models on physical properties of small molecules. Uses user provided custom feature vector and machine learning to train said models.
ML Predict: Regression using Feature Input predicts property from a pretrained neural network model. Needs molecule database, feature vector(same length as training data), and neural network model dataset as input (trained using the previous floe).
Data Processing of Small Molecule for ML Model Building analyzes and preprocesses data for training ML(machine learning) models. Needs molecule database and response value (either for classification or regression).
Old Floes in this package
ML Regression Model Building using Fingerprints for Small Molecules (previously Neural Network Based Regression Model Building) trains multiple Neural Network (Full or Probabilistic) Regression models on physical properties of small molecules. Uses cheminformatics and machine learning to train said models.
ML Regression using Fingerprints for Small Molecules (previously Physical Property Prediction for Small Molecules using Machine Learning) predicts property from a pretrained neural network model. Needs molecule database and neural network model dataset as input (trained using the previous floe).
Solubility Prediction for Small Molecules using Machine Learning and Cheminfo Fingerprints predicts solubility in log uM using neural network based machine learning. Needs molecule database as input.
v0.9.2 September 2022
Minor Bug Fixes
This package is built using
OpenEye-orionplatform==4.4.0,OpenEye-toolkits==2022.1.1, andOpenEye-Snowball==0.24.0..Bug fixed on Physical Property Prediction for Small Molecules using Machine Learning
Model and Record ID in Neural Network Based Regression Model Building renamed for user convenience.
Better use case handling for wrong inputs.
v0.9.0 July 2022
General Notice
This is the first release of Model Building package with package version 0.9.0.
This package is built using
OpenEye-orionplatform==4.4.0,OpenEye-toolkits==2022.1.1, andOpenEye-Snowball==0.24.0..All floes have a new brief parameter and floe categories.
Floes in this package
Neural Network Based Regression Model Building trains multiple Neural Network (Full or Probabilistic) Regression models on physical properties of small molecules. Uses cheminformatics and machine learning to train said models.
Physical Property Prediction for Small Molecules using Machine Learning predicts property from a pretrained neural network model. Needs molecule database and neural network model dataset as input (trained using the previous floe).
Solubility Prediction for Small Molecules using Machine Learning and Cheminfo Fingerprints predicts solubility in log uM using neural network based machine learning. Needs molecule database as input.