ChemProps

ChemProps standardizes polymer names and maps them to canonical identifiers and SMILES to enable consistent indexing in materials informatics.

Key Features:

• RESTful API: Exposes programmatic access for integration with external databases and services.
• Multi-algorithm mapping: Uses a multi-algorithm approach to resolve non-uniform expressions of polymer names.
• Score-based ranking: Assigns weight factors to each algorithm to compute scores for candidate chemical names.
• Weight optimization: Optimizes weight factors to maximize the score difference between the ground-truth polymer name and alternative candidates.
• Cross-validation: Applies ten-fold validation on 160 training data points to assess model robustness.
• Reported test performance: Achieved 100% test accuracy on 54 test data points.
• Dynamic weights: Updates weight factors as the underlying database expands to adapt scoring over time.
• Name-to-SMILES translation: Provides standardized name-to-SMILES mappings for chemical representation.
• Duplicate detection: Identifies and facilitates removal of duplicate polymer entries in databases.
• Property auto-population: Enables auto-population of polymer properties using standardized identifiers.
• Extensible updates: Supports incorporation of new polymer knowledge to broaden and deepen the database content.

Scientific Applications:

• Polymer name standardization: Harmonizes polymer nomenclature across heterogeneous data resources to improve indexing consistency.
• Database curation: Detects and helps remove duplicate records to improve database quality.
• Structure representation: Maps polymer names to SMILES for consistent molecular representation in analyses.
• Materials informatics integration: Facilitates interoperability between polymer databases for downstream informatics workflows.
• Property annotation: Supports automated annotation and population of polymer properties for research and database enrichment.

Methodology:

Uses a multi-algorithm mapping approach that assigns weight factors to algorithms to score candidate names, optimizes those weights to maximize score separation between ground-truth and alternatives, validates performance with ten-fold cross-validation on 160 training points and reports outcomes on 54 test points, and updates weights as the database expands.