Multi-Instance Learning Approach to Predictive Modeling of Catalysts Enantioselectivity

Abstract

Here, we report an application of the multi-instance learning approach to predictive modeling of enantioselectivity of chiral catalysts. Catalysts were represented by ensembles of conformations encoded­ by the pmapper physicochemical descriptors capturing stereoconfiguration of the molecule. Each catalyzed chemical reaction was transformed to a condensed graph of reaction for which ISIDA fragment descriptors were generated. This approach does not require any conformations’ alignment and can potentially be used for a diverse set of catalysts bearing different scaffolds. Its efficiency has been demonstrated in predicting the selectivity of BINOL-derived phosphoric acid catalysts in asymmetric thiol addition to N-acylimines and benchmarked with previously reported models.

Predicting Highly Enantioselective Catalysts Using Tunable Fragment Descriptors**

“…The disadvantage of such methods is the requirement of costly quantum chemical calculations and, in the case of grid‐based methods, the necessity of alignment of core structures. Although several 3D structure‐based methods that do not require alignment have also been described in the literature, their performances, especially in extrapolation, are limited [3, 21–23] …”

“…Although several 3D structure-based methods that do not require alignment have also been described in the literature, their performances, especially in extrapolation, are limited. [3,[21][22][23] In contrast, 2D descriptors, such as fragment counts or binary fingerprints, also represent general structural features, albeit often implicitly, and allow for avoiding costly calculations, as they are derived directly from a 2D representation of a molecule. [24][25][26][27] Using 2D descriptors, therefore, has a clear advantage in speed.…”

Multi-Instance Learning Approach to the Modeling of Enantioselectivity of Conformationally Flexible Organic Catalysts

“…Each generated catalyst conformer was encoded by pmapper descriptors 25 representing various combinations of 3D pharmacophore quadruplets. 20,21 In this work, instead of pharmacophore features used in our early study 19 , we used quadruplets and triplets enumerating ensembles of individual atoms and/or centers of 5-and 6-membered aromatic rings. Notice that application of atoms triplets significantly reduces the number of descriptors, and related models perform similarly to those built on stereosensitive atoms quadruplets.…”

“…Recently, we have reported an alternative approach that used Multi-Instance Learning (MIL) algorithms 18 accounting for all low-energy conformers of catalysts encoded by alignment-independent 3D pmapper descriptors 19 in combination with the compact representation of chemical reactions by their Condensed Graph of Reaction 22 . This approach provided with models performance similar to Zahrt et al 10 .…”

“…This approach provided with models performance similar to Zahrt et al 10 . Unfortunately, the short communication format of publication 19 did not allow us to describe the modeling workflow in detail.…”