Physical Pooling Functions in Graph Neural Networks for Molecular Property Prediction

RWTH Publications (RWTH Aachen)

Abstract

Graph neural networks (GNNs) are emerging in chemical engineering for the end-to-end learning of physicochemical properties based on molecular graphs. A key element of GNNs is the pooling function which combines atom feature vectors into molecular fingerprints. Most previous works use a standard pooling function to predict a variety of properties. However, unsuitable pooling functions can lead to unphysical GNNs that poorly generalize. We compare and select meaningful GNN pooling methods based on physical knowledge about the learned properties. The impact of physical pooling functions is demonstrated with molecular properties calculated from quantum mechanical computations. We also compare our results to the recent set2set pooling approach. We recommend using sum pooling for the prediction of properties that depend on molecular size and compare pooling functions for properties that are molecular size-independent. Overall, we show that the use of physical pooling functions significantly enhances generalization.

Authors 7

1. Artur M. Schweidtmann Aachen

   RWTH Aachen University · Delft University of Technology

   Affiliation as printed

   Current address: Delft University of Technology , Department of Chemical Engineering , Van der Maasweg 9 , Delft 2629 HZ , The Netherlands

   RWTH Aachen University , Process Systems Engineering (AVT.SVT) , Forckenbeckstr. 51 , 52074 Aachen , Germany

2. Jan G. Rittig Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University , Process Systems Engineering (AVT.SVT) , Forckenbeckstr. 51 , 52074 Aachen , Germany

3. Jana M. Weber

   Delft University of Technology

   Affiliation as printed

   Delft University of Technology , Delft Bioinformatics Lab , Intelligent Systems , TU Delft , 2628 XE Delft , The Netherlands

4. Martin Grohe Aachen Lehrstuhl für Informatik 7

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University , Lehrstuhl für Informatik 7 , Ahornstr. 55 , 52074 Aachen , Germany

5. Manuel Dahmen

   Forschungszentrum Jülich

   Affiliation as printed

   Forschungszentrum Jülich GmbH , Institute of Energy and Climate Research IEK-10 -Energy Systems Engineering , Wilhelm-Johnen-Str. , 52425 Jülich , Germany

6. Kai O. Leonhard Aachen Institute of Technical Thermodynamics

   RWTH Aachen University · FH Aachen

   Affiliation as printed

   RWTH Aachen University , Institute of Technical Thermodynamics , Schinkelstr. 8 , 52062 Aachen , Germany

7. Alexander Mitsos