Multimodal Graph-based Transformer Framework for BiomedicalRelation Extraction

 Abstract

task：Protein-Protein Interaction task

introduced a novel framework that enables the model to learn multi-omnics biological information about entities (proteins) with the help of additional multi-modal cues like molecular structure.

devise a generalized and optimized graph based multi-modal learning mechanism that utilizes the GraphBERT model

1 Introduction

The main contributions：

utilized protein atomic structural information while identifying the protein interactions

developed a generalized modality-agnostic approach

2 Proposed Method

four main components:

(1) Multi-modal Graph Constructor ,

(2) Multi -modal Graph Fusion

(3) Multi-modal Graph En coder

(4) PPI Predictor .

Problem Statement:

a biomedical input text：

S = { w 1 , w 2 , . . . , w n }

a pair of protein mentions：

p 1 , p 2 ∈ S

predict:

‘interact’ or ‘non-interact’

2.1 Multi-modal Graph Constructor

Textual Graph Constructor：

constructs the graph by considering the textual content that aims to capture the lexical and contextual informa tion present in the input

Protein Structure Graph Constructor：

exploits the atomic structure (3D PDB structure) of the protein molecules to build the graph.

2.2 Multi-modal Graph Fusion

2.3 Multi-modal Graph Encoder

2.4 PPI Predictor

3 Datasets and Experimental Analysis

 

 

 

4 Error Analysis