SpectraFP: a new spectra-based descriptor to aid in cheminformatics, molecular characterization and search algorithm...

Area of Science:

• Spectroscopy and Cheminformatics
• Computational Chemistry
• Machine Learning in Chemistry

Background:

• Digitalizing spectroscopic data, particularly ¹³C NMR chemical shifts, is crucial for computational chemistry and cheminformatics.
• Existing methods may not fully capture the nuances of spectral data or correct for fluctuations.
• There is a need for robust, versatile descriptors for spectroscopic information.

Purpose of the Study:

• To introduce SpectraFP, a novel spectra-based descriptor for digitalizing ¹³C NMR chemical shifts and other spectroscopic data.
• To demonstrate the applicability of SpectraFP in predicting functional groups using machine learning (ML) models.
• To develop a structure search algorithm leveraging SpectraFP for similarity comparisons against experimental spectra.

Main Methods:

• Development of the SpectraFP algorithm, a binary fingerprint vector designed to represent spectroscopic data.
• Construction and validation of ML models for predicting six functional groups, adhering to OECD principles (internal/external validation, applicability domains, mechanistic interpretation).
• Implementation of a similarity search algorithm using various metrics (Tanimoto, geometric, arithmetic, Tversky) and incorporating additional parameters.

Main Results:

• ML models for functional group prediction showed high goodness-of-fit (MCC between 0.626-0.917, J between 0.812-0.961).
• SHAP analysis confirmed that model decisions were mechanistically interpretable and aligned with expected chemical shifts.
• The similarity search algorithm demonstrated high performance speed and flexibility with additional variable incorporation.

Conclusions:

• SpectraFP is a valuable tool for digitalizing spectroscopic data, enabling robust ML applications in cheminformatics.
• The developed ML models and search algorithm show significant potential for chemical structure elucidation and analysis.
• Open-source availability of databases and algorithms facilitates further research and application in the field.