PEDL

PEDL (PPA Extraction with Deep Language) is an approach to predict protein-protein associations (PPAs) from text, addressing the critical need for an up-to-date and comprehensive resource of functional PPAs that underlie biological signaling pathways. Recognizing the limitations of current extraction methods that heavily depend on scarce manually labeled data, PEDL leverages deep language models in conjunction with distant supervision, thereby gaining access to a substantially larger pool of training data than methods restricted to manually annotated annotations.

This methodology allows PEDL to not only predict PPAs between two proteins but also identify the specific text spans within biomedical literature that support the existence of these associations. Evaluation of PEDL across three different datasets demonstrated its superior performance in identifying PPAs and pinpointing supporting text spans, outperforming a recent state-of-the-art model across all metrics.