Improving Molecular Graph Neural Network Explainability
          with Orthonormalization and Induced Sparsity

              Ryan Henderson 1 Djork-Arne Clevert 1 Floriane Montanari 1

             Abstract                 However, the opaque nature of artificial neural networks
                                 has long been a stumbling block for wider adoption. In
   Rationalizing which parts of a molecule drive the      particular, the difficulty in extracting a straightforward ra-
   predic ...