Decoupled coordinates for machine learning-based molecular fragment linking

Recent developments in machine-learning based molecular fragment linking have demonstrated the importance of informing the generation process with structural information specifying the relative orientation of the fragments to be linked. However, such structural information has not yet been provided in the form of a complete relative coordinate system. Mathematical details for a decoupled set of bond lengths, bond angles and torsion angles are elaborated and the coordinate system is demonstrated to be complete. Significant impact on the quality of the generated linkers is demonstrated numerically. The amount of reliable information within the different types of degrees of freedom is investigated. Ablation studies and an information-theoretical analysis are performed. The presented benefits suggest the application of a complete and decoupled relative coordinate system as a standard good practice in linker design.