Developing a Hybrid Molecular Representation Combining Chemical Structure and MIR Spectral Data: A LogP Prediction...

Area of Science:

• Cheminformatics
• Spectroscopy
• Quantitative Structure-Activity Relationship (QSAR)

Background:

• Molecular fingerprints are crucial for representing chemical structures in computational modeling.
• Integrating experimental data like mid-infrared (MIR) spectra into molecular representations can enhance predictive models.
• Current methods for logP prediction often rely solely on structural information.

Purpose of the Study:

• To develop and evaluate a novel hybrid molecular fingerprint combining structural and MIR spectral data.
• To assess the performance of this fingerprint in a logP prediction task.
• To explore the feasibility of incorporating MIR spectral information into Quantitative Structure-Activity Relationship (QSAR) workflows.

Main Methods:

• A 101-bit binary hybrid molecular fingerprint was created, encoding both structural substructures and MIR absorption bands.
• Support Vector Regression (SVR) was employed to predict logP values using the hybrid fingerprint.
• Performance was benchmarked against traditional structure-based fingerprints and existing logP prediction tools.

Main Results:

• The hybrid fingerprint achieved a Root Mean Square Error (RMSE) of 1.443 in logP prediction.
• Traditional fingerprints (Morgan, MACCS) and descriptor-based models showed lower RMSEs.
• Commercial and open-source logP tools also outperformed the hybrid fingerprint in this specific task.

Conclusions:

• The proposed hybrid fingerprint, though modest in predictive accuracy for logP, provides a novel, interpretable, and computationally efficient method.
• It successfully demonstrates the integration of MIR spectral data into cheminformatics modeling.
• This work lays the groundwork for developing advanced spectrum-informed molecular representations for QSAR studies.