Download Schrodinger DeepAutoQSAR Tutorials
Schrödinger DeepAutoQSAR Tutorials provides practical example datasets, pretrained models, prediction files, and workflow scripts for exploring automated machine learning-based molecular property prediction. DeepAutoQSAR is an automated and scalable machine learning solution that allows users to predict molecular properties from chemical structures and develop high-performing quantitative structure–activity relationship (QSAR) and quantitative structure–property relationship (QSPR) models. Its automated, supervised learning pipeline is designed for both novice and experienced researchers, supporting the training and application of predictive machine learning models across a wide range of chemical and molecular modeling applications.
Key Capabilities of DeepAutoQSAR
Automated Machine Learning Workflows
DeepAutoQSAR streamlines the model-building process by automatically generating molecular descriptors and fingerprints, training models using multiple machine learning architectures, and evaluating model performance. This automated workflow makes QSAR and QSPR model development more accessible to both experienced computational scientists and users who are new to machine learning.
Project-Specific Descriptors
Users can provide their own molecular descriptors in CSV format and use them alongside, or instead of, descriptors generated by DeepAutoQSAR. This flexibility allows the platform to be adapted to specialized research projects and applications beyond conventional small-molecule modeling, including polymers, organic electronics, catalysis, and other areas of molecular science.
Model Optimization and QSAR Best Practices
DeepAutoQSAR incorporates established QSAR/QSPR modeling best practices to reduce the risk of overfitting and misleading performance estimates. Automated model evaluation and optimization help identify effective model architectures while providing a more reliable assessment of predictive performance.
Model Confidence and Applicability Domain
DeepAutoQSAR provides uncertainty or confidence estimates together with molecular property predictions. These estimates help researchers assess the reliability of predictions, particularly when candidate molecules fall outside or near the boundaries of the chemical space represented by the training dataset.
Visualization and Analysis in Maestro
DeepAutoQSAR results can be visualized and analyzed in Maestro, allowing researchers to examine model metrics, performance plots, and predictions. Atomic contribution visualizations can also highlight which parts of a molecular structure contribute to the predicted target property, providing useful insights for chemical design and the generation of new molecular ideas.
Scalable Machine Learning for Small and Large Datasets
DeepAutoQSAR supports different machine learning approaches according to dataset size and modeling requirements. Classical machine learning methods, including boosted-tree approaches, can be applied to smaller datasets, while graph neural networks and other modern deep learning methods enable the development of large-scale QSAR/QSPR models.
Applications in Computational Chemistry and Drug Discovery
DeepAutoQSAR can be used for a wide range of molecular property prediction and QSAR/QSPR applications. Its automated workflows and scalable machine learning methods make it suitable for areas such as drug discovery, medicinal chemistry, ADMET prediction, molecular property prediction, cheminformatics, chemical biology, and computational chemistry.
By combining automated descriptor generation, multiple machine learning architectures, model optimization, uncertainty estimation, and visualization capabilities, DeepAutoQSAR provides a comprehensive workflow for developing and applying predictive models to chemical datasets.
Who Should Use These Tutorials?
These tutorials are suitable for computational chemists, medicinal chemists, cheminformatics researchers, molecular modelers, drug discovery scientists, QSAR/QSPR researchers, and students interested in applying machine learning to chemical structures and molecular properties.
The included datasets, pretrained models, prediction examples, and workflow scripts provide a practical starting point for learning how DeepAutoQSAR can automate the development and application of predictive machine learning models in computational chemistry and drug discovery.
Schrodinger DeepAutoQSAR Tutorials details:
- Supported operating systems: Windows, Linux, macOS
- File size: 3.24 GB
- Note: GPU machine learning applications such as Active Learning Glide and DeepAutoQSAR on GPU can only be run on Linux.
