Fig. 1
From: Learning motif features and topological structure of molecules for metabolic pathway prediction
The structure of MotifMol3D framework for metabolic pathway prediction. The framework comprises four sections: (1) The model takes the SMILES representation of small molecules as input; (2) In the molecular feature extraction module, a heterogeneous motif graph network containing motifs and molecular nodes is established to learn motif features. Additionally, global features of molecules are obtained by introducing a graph attention network; (3) In the multi-label classification module, the features from the output layer are merged and fed into a fully connected (FC) layer, which is then trained with labels from the training set. Finally, XGBoost is employed to predict the molecular pathway categories; (4) The model's ultimate output indicates participation (1) or non-participation (0) in each of the 11 metabolic pathway categories