Combining Group-Contribution Concept and Graph Neural Networks Toward Interpretable Molecular Property Models

Abstract

Quantitative structure–property relationships (QSPRs) are important tools to facilitate and accelerate the discovery of compounds with desired properties. While many QSPRs have been developed, they are associated with various shortcomings such as a lack of generalizability and modest accuracy. Albeit various machine-learning and deep-learning techniques have been integrated into such models, another shortcoming has emerged in the form of a lack of transparency and interpretability of such models. In this work, two interpretable graph neural network (GNN) models (attentive group-contribution (AGC) and group-contribution-based graph attention (GroupGAT)) are developed by integrating fundamentals using the concept of group contributions (GC). The interpretability consists of highlighting the substructure with the highest attention weights in the latent representation of the molecules using the attention mechanism. The proposed models showcased better performance compared to classical group-contribution models, as well as against various other GNN models describing the aqueous solubility, melting point, and enthalpies of formation, combustion, and fusion of organic compounds. The insights provided are consistent with insights obtained from the semiempirical GC models confirming that the proposed framework allows highlighting the important substructures of the molecules for a specific property.

Extrapolative prediction of small-data molecular property using quantum mechanics-assisted machine learning

“…The matrix B 00 N M 00 denotes the interaction terms for N molecules, where each row corresponds to b″ in (9).…”

“…Materials science has greatly benefited from advancements in machine learning (ML) and deep learning (DL) techniques [1][2][3][4][5][6] . These techniques have revolutionized the prediction of molecular properties, leveraging traditional computational approaches, such as the group contribution (GC) method [7][8][9] , quantitative structureactivity/property relationship (QSAR/QSPR) method [10][11][12][13] , quantum mechanics (QM), and molecular dynamics (MD) calculations [14][15][16][17][18][19][20][21][22][23][24][25][26] . Graph neural networks (GNNs) have emerged as a promising DLbased method for property prediction by embedding molecular structures in a graph architecture 27,28 .…”

Group Contribution Revisited: The Enthalpy of Formation of Organic Compounds with “Chemical Accuracy” Part V

“…For the purpose of the evaluation of the enthalpy of formation ∆H f of organic molecules from their molecular structure, the Group Contribution (GC) approach is one of the most important and widely applied methods [1][2][3][4][6][7][8][9][10][11][12][13][14][15]. The original GC method is based on the assumption that a molecule can be decomposed into molecular fragments which are in essence mutually independent, and the molecular property of interest is the sum of the individual properties of the molecular fragments.…”

Group Contribution Revisited: The Enthalpy of Formation of Organic Compounds with “Chemical Accuracy” Part IV

“…In the course of time, a larger number of further studies employing the GC methodology to evaluate the heat of formation of organic molecules have been reported, including the works by Benson and co-workers [2], Joback and Reid [3] and Gani et al [4,5]. Recently, a study relevant in this context and employing neural networks was also reported [6].…”