Biodegradation QSAR Dataset with 10 Molecular Classes

QSAR-Biodeg contains molecular descriptor data for predicting chemical biodegradability, derived from the original QSAR biodegradation dataset. The seed_3_nrows_2000_nclasses_10_ncols_100_stratify_True parameters suggest a structured subset of 2000 instances, 100 features, and 10 balanced target classes. This version is hosted on the OpenML platform for reproducible machine learning experiments.

Use Cases

• Predicting the biodegradation class (e.g., ready or not ready biodegradable) from 100 molecular descriptor columns using random forests.
• Identifying the most influential molecular features among the 100 descriptors for biodegradability via feature importance analysis.
• Training a multi-class classifier like XGBoost on the 10 target classes using the stratified 2000-row sample.
• Conducting dimensionality reduction (e.g., PCA) on the 100-column feature space to visualize chemical clusters.

Strengths

• Stratified sampling ensures balanced representation across the 10 target classes.
• Contains 100 molecular descriptors providing a detailed feature set for modeling.
• The 2000-row sample size is suitable for training and validating machine learning models.

Limitations

• The specific 100 molecular descriptors are unknown, complicating domain interpretation.
• As a derived subset, it may not represent the full chemical diversity of the original dataset.
• Lack of metadata on source molecules and experimental methods limits reproducibility for chemists.

Provenance

Source

OpenML platform, derived from the QSAR Biodegradation Dataset.

Collection Method

Algorithmic sampling (seed=3) from a larger QSAR dataset to create a balanced subset.