Skip to main content

Table 3 Other model architecture differences

From: Building attention and edge message passing neural networks for bioactivity and physical–chemical property prediction

ModelPre-message passingUpdatePre-readout
MPNN
AMPNN
NA\(h_{v}^{{\left( {t + 1} \right)}} = {\text{GRU}}\left( {m_{v}^{\left( t \right)} ,h_{v}^{\left( t \right)} } \right)\)NA
EMNN\(e_{vw}{^{\prime}} = f_{NN}^{emb} \left( {\left( {e_{vw} ,h_{v}^{\left( 0 \right)} ,h_{w}^{\left( 0 \right)} } \right)} \right)\)\(h_{vw}^{{\left( {t + 1} \right)}} = {\text{GRU}}\left( {m_{vw}^{\left( t \right)} ,h_{vw}^{\left( t \right)} } \right)\)\(h_{v}^{\left( K \right)} = \mathop \sum \limits_{w \in N\left( v \right)} h_{vw}^{\left( K \right)}\)