MOSES

MOSES benchmarks molecular generative models by providing standardized datasets, model implementations, and evaluation metrics to assess the quality and diversity of generated molecules for drug discovery and generative chemistry.

Key Features:

• Standardization of Research: Provides a unified framework and standardized datasets for training and comparing molecular generative models.
• Model Implementations: Includes implementations of several popular molecular generation models for direct comparison.
• Comprehensive Metrics: Offers a set of metrics to assess the quality and diversity of generated molecular structures.
• Training and Testing Datasets: Distributes curated training and testing datasets as benchmarks for reproducible evaluation.
• Facilitation of Model Comparison: Enables ranking and comparison of generative models through standardized evaluation procedures.

Scientific Applications:

• Virtual Screening: Generated molecular structures can be used for virtual screening to identify potential drug candidates.
• Semi-Supervised Learning: Novel structures produced by generative models can augment datasets for semi-supervised predictive models.

Methodology:

Implements and compares several molecular generation models using curated standardized datasets to produce benchmarking reference points.