Identifying Protein-Protein Interaction using Tree-Transformers and Heterogeneous Graph Neural Network

For a better understanding of the underlying biological mechanisms, it is crucial to identify reciprocity between proteins. Often, extracting such interactions proteins from biomedical articles faces challenges due complex sentence structure textual information sources. Most prominent previous works have applied additional hand-crafted features for protein-protein interaction task. In this work, we utilized two tree-structured attention-based neural network models along with heterogeneous graph approach perform We suggest that proposed model preserves syntactic as well semantic text. The experimental results demonstrate even without using any feature extraction techniques, achieves significant performance boosts when on five standard benchmark corpora compared works.