CIMtools
CIMtools implements specialized cross-validation strategies for cheminformatics QSPR models by applying 'transformation-out' and 'solvent-out' cross-validation to reduce optimistic bias and improve performance estimation on unseen reaction types and solvent conditions.
Key Features:
- Transformation-out cross-validation: Excludes specific transformation classes from training to test model performance on unseen reaction types.
- Solvent-out cross-validation: Excludes specific solvents from training to evaluate model generalizability across different solvent conditions.
- Bias mitigation for k-fold CV: Addresses optimistic bias associated with conventional k-fold cross-validation by incorporating the nature of chemical objects into validation.
- QSPR modeling for reactions: Targets quantitative structure–property relationship (QSPR) models focused on reaction rate prediction and reaction kinetics.
- ISIDA Fragmentor integration: Provides a Python wrapper for ISIDA Fragmentor for fragment-based descriptor generation.
- RDtool atom-to-atom mapper integration: Provides a Python wrapper for RDtool atom-to-atom mapper for atom-mapping of reactions.
Scientific Applications:
- Rate constant prediction (bimolecular elimination): Predicts rate constants for bimolecular elimination reactions.
- Rate constant prediction (nucleophilic substitution): Predicts rate constants for nucleophilic substitution reactions.
- Rate constant prediction (Diels–Alder cycloaddition): Predicts rate constants for Diels–Alder cycloaddition reactions.
- Reaction kinetics and mechanistic studies: Provides more reliable model evaluation for studies of reaction kinetics and mechanisms.
- Generalization testing: Enables assessment of model performance on novel structural transformations and reactions under new conditions.
Methodology:
Applies transformation-out and solvent-out cross-validation to QSPR models, implements exclusion of specific transformation classes or solvents from training, and interfaces via a Python wrapper with ISIDA Fragmentor and RDtool atom-to-atom mapper.
Topics
Details
- License:
- GPL-3.0
- Tool Type:
- library
- Operating Systems:
- Mac, Linux, Windows
- Programming Languages:
- Python
- Added:
- 3/19/2021
- Last Updated:
- 4/11/2021
Operations
Publications
Rakhimbekova A, Akhmetshin T, Minibaeva G, Nugmanov R, Gimadiev T, Madzhidov T, Baskin I, Varnek A. Cross-validation strategies in QSPR modelling of chemical reactions. SAR and QSAR in Environmental Research. 2021;32(3):207-219. doi:10.1080/1062936x.2021.1883107. PMID:33601989.