Knowledge of the Ancestors: Intelligent Ontology-Aware Annotation of Biological Literature Using Semantic Similarity

EasyChair Preprint 8966

Abstract

Natural language processing models have emerged as a solution to manual curation for fast and automated annotation of literature with ontology concepts. Deep learning architectures have particularly been employed for this task due to increased accuracy over traditional machine learning techniques. One of the greatest limitations in prior work is that the architectures do not use the ontology hierarchy while training or making predictions. These models treat ontology concepts as if they were independent entities while ignoring the semantics and relationships represented in the ontology. Here, we present deep learning architectures for ontology-aware models that use the ontology hierarchy for training and predicting ontology concepts for pieces of text.

We explore the choice of three embeddings - CRAFT, GloVe, and ELMo to understand the impact on prediction performance. We evaluate our models using F-1 and Jaccard semantic similarity and show that our ontology aware models can result in 2% - 10% (depending upon choice of embedding) improvements over a baseline model that doesn't use ontology hierarchies.

Keyphrases: Gene Ontology, automated annotation, biocuration, deep learning

@booklet{EasyChair:8966,
  author    = {Pratik Devkota and Somya Mohanty and Prashanti Manda},
  title     = {Knowledge of the Ancestors: Intelligent Ontology-Aware Annotation of Biological Literature Using Semantic Similarity},
  howpublished = {EasyChair Preprint 8966},
  year      = {EasyChair, 2022}}