Alvascience recently released a new major version of our tool, alvaModel, our software tool to build, compare, and deploy QSAR/QSPR Models (https://www.alvascience.com/alvamodel/).
New Modeling Capabilities
Regression and Classification Models
– Decision Tree (Regression and Classification)
– Random Forest (Regression and Classification)
– Consensus Classification Model (weighted on reliability)
Customizable Model Parameters
Users can now define sets of custom parameters for:
– K-Nearest Neighbors (KNN)
– Support Vector Machine (SVM)
– Consensus (Classification)
– Decision Tree
– Random Forest
Applicability Domain
– New method: Bounding Box
Similarity Measures
– Dice and Cosine distances are now supported
Enhanced Evaluation Metrics
New Classification Scores
– AUROC (Area Under the Receiver Operating Characteristic Curve)
– F1 Score
– Matthews Correlation Coefficient (MCC)
– Cohen’s Kappa
Improved Visualization and User Interface
Model Comparison Features:
– New interface for comparing multiple models
– Display of 95% confidence intervals for scores
– Automatic highlighting of best/worst scores
New comparison charts:
– Simultaneous Confidence Interval Plot
– Radar Plot
Classification Model View Enhancements
– New “Reliability” column
– Color-coded indicators for:
– Correct predictions
– Reliability
– Applicability Domain (A.D.)
– Colored Confusion Matrix with molecule filtering capability
Workflow and Usability Improvements
– Prediction of external datasets without the need for alvaRunner
– New “Open Recent Project” menu
– Batch model saving in automatic model generation
– Improved dataset management:
– Move or copy molecules between datasets
– Rename external variable columns
– Contextual popup menus for dataset and model tables
Charts and Visualization Enhancements
– Unified chart toolbar for easy selection of chart types and settings
– For regression models: R² and RMSE are now displayed in plot legends
– For classification models: ROC Curve and Precision–Recall Curve available