Enhancing the Interpretability of Asymmetric Catalysis CoMFA through PLS (n = 1) with Contribution Map Merging and Spatial Aggregation

Although CoMFA/PLS is a powerful QSAR tool, its interpretation can be complicated by multicollinearity. To enhance its interpretability in asymmetric catalysis, we developed a highly predictive model for enantioselectivity in Pd-catalyzed asymmetric fluorination. This was achieved stably by integrating single-latent-variable PLS (n = 1) models with contribution map merging and spatial aggregation. Overlap analysis between the aggregated patterns and molecular structures revealed a strong correlation (Pearson correlation coefficient R ≈ 0.8–0.9) between the sum of the coefficients in the overlap region and the experimental reaction barrier (ΔΔG‡). The resulting spatial patterns were also consistent with those of known chemical findings. This methodology enables the extraction of stable, interpretable, and highly predictive spatial contribution patterns. This approach is expected to facilitate rational molecular design by providing reliable mechanistic insights.