Hierarchical multi-label classification based on LSTM network and Bayesian decision theory for LncRNA function...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Long noncoding RNAs (lncRNAs) are crucial regulators in cellular processes and disease.
• Most lncRNA functions remain uncharacterized, hindering biological understanding.
• Existing computational methods often overlook the hierarchical structure of gene ontology (GO) annotations.

Purpose of the Study:

• To develop an accurate computational method for predicting lncRNA functions.
• To address the limitations of current methods by incorporating directed acyclic graph (DAG) structures in GO annotations.
• To treat lncRNA function annotation as a hierarchical multilabel classification problem.

Main Methods:

• Proposed a novel method, HLSTMBD, for hierarchical multilabel classification with DAG-structured labels.
• Implemented HLSTMBD using a long-short term memory (LSTM) network and a hierarchical constraint method (DAGLabel).
• Utilized a mathematical model based on Bayesian decision theory for algorithm implementation.

Main Results:

• HLSTMBD demonstrated efficient and accurate performance in lncRNA function prediction.
• The method successfully handled DAG-structured labels inherent in GO annotations.
• Comparative analysis showed superior results against state-of-the-art algorithms on GOA-lncRNA datasets.

Conclusions:

• HLSTMBD offers an effective approach for computational lncRNA function annotation.
• The method advances the prediction of gene functions by leveraging hierarchical and structured label information.
• This work contributes to a better understanding of lncRNA roles in biological processes and disease.