We present an approach to reduce this computational cost substantially, based on the partitioning of the molecule into geometrically separated torsional groups, with the dependence of the intramolecular energy and atomic point charges and dependent degrees of freedom on molecular conformation being computed as a linear combination of the contributions of these groups. This can lead to large savings in computational cost without a significant impact on accuracy, as demonstrated in the cases of N-acetyl-para-aminophenol (paracetamol) and methyl 4-hydroxybenzoate (methyl paraben). The approach is also applied successfully to two larger molecules, benzyl [4-(4-methyl-5-[(4-methylphenyl)sulfonyl]-1,3-thiazol-2-yl)phenyl]carbamate (molecule XX from the fifth CSP blind test) and (2S)-2-[4-(3-fluorobenzyloxy)benzylamino]propionamide (safinamide), for which we conduct the first reported CSP study. In both cases, the use of torsional groups results in over 99% reduction in computational cost, which enables the generation of an initial CSP landscape with high-quality structures found within the standard cutoff of 20 kJ mol-1 for progression to refinement.
Read full abstract