OTMol: Robust Molecular Structure Comparison via Optimal Transport.

Area of Science:

• Computational chemistry
• Structural bioinformatics
• Machine learning in chemistry

Background:

• Root-mean-square deviation (RMSD) is crucial for molecular structural similarity assessment.
• Traditional RMSD methods struggle with atom ordering, cluster configurations, and chirality.
• Existing alignment algorithms often fail to generalize across diverse chemical systems.

Purpose of the Study:

• To develop a robust and generalizable molecular alignment method.
• To overcome limitations of traditional RMSD calculations, including atom correspondence issues.
• To leverage intrinsic molecular information for accurate structural comparisons.

Main Methods:

• Formulated molecular alignment as a fused supervised Gromov-Wasserstein (fsGW) optimal transport problem.
• Utilized intrinsic geometric and topological relationships within molecules for data-driven matching.
• Ensured one-to-one atom mappings to maintain molecular integrity.

Main Results:

• OTMol achieves low RMSD values across diverse systems like ATP, imatinib, lipids, peptides, and water clusters.
• The method preserves crucial chemical features, including molecular chirality and bond connectivity.
• OTMol demonstrates computational efficiency and avoids erroneous many-to-one alignments in clusters.

Conclusions:

• Optimal transport theory provides a powerful framework for molecular alignment.
• OTMol offers a principled, data-driven approach superior to heuristic methods.
• This method advances structural comparison in cheminformatics and molecular modeling.