MolRepp

MolRepp standardizes evaluation of deep representation learning models for molecular property prediction to enable rigorous, reproducible comparisons across models and molecular representations.

Key Features:

• Unified Framework: Consolidates 16 deep representation models and four molecular representations into a single evaluation environment.
• Extensive Benchmarking: Re-evaluates 16 state-of-the-art deep representation models across 12 common benchmark datasets.
• Hyperparameter Optimization: Performs exhaustive hyperparameter optimization with over 12.5 million experiments across methods and datasets.
• Performance Metrics and Rankings: Reports comparative results with CMPNN first in five of twelve tasks (average rank 1.75), ECC top in three classification tasks (average rank 2.71), and MAT top in three regression tasks (average rank 2.6).
• Reproducible Evaluation: Implements standardized evaluation procedures to enable fair, reproducible comparisons of molecular property prediction algorithms.

Scientific Applications:

• Comparative model evaluation: Supports comparative analysis of deep representation learning models for molecular property prediction in bioinformatics and cheminformatics.
• Model selection and development: Aids selection and development of molecular property prediction algorithms using validated benchmarking and hyperparameter-optimized results.

Methodology:

Integrates 16 deep representation models with four molecular representations and performs hyperparameter optimization using over 12.5 million experiments across 12 benchmark datasets.