A deep neural network based hierarchical multi-label classification method.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene function prediction is crucial due to increasing biological data.
• Gene Ontology (GO) annotations form a directed acyclic graph (DAG), posing challenges for standard hierarchical multi-label classification (HMC) algorithms.
• Existing HMC algorithms struggle with DAG structures, leading to low accuracy in gene function prediction.

Purpose of the Study:

• To propose a novel HMC algorithm tailored for DAG structures, specifically for GO-based gene function prediction.
• To address the limitations of existing algorithms in handling the hierarchical constraints of DAGs.
• To improve the accuracy and applicability of HMC methods in bioinformatics.

Main Methods:

• Developed the C2AE-DAGLabel algorithm, integrating a Canonical Correlated AutoEncoder (C2AE) model.
• Designed a DAGLabel algorithm to enforce hierarchical constraints within the DAG structure.
• Evaluated the algorithm using human gene data annotated with GO.

Main Results:

• The C2AE-DAGLabel algorithm effectively handles the hierarchical constraints of GO DAGs.
• Experimental results demonstrate superior performance compared to state-of-the-art algorithms.
• The proposed method significantly improves accuracy in DAG-based HMC for gene function prediction.

Conclusions:

• The C2AE-DAGLabel algorithm provides an effective solution for gene function prediction on GO DAGs.
• The DAGLabel component successfully enforces hierarchical consistency, enhancing classification accuracy.
• This advancement offers a more reliable tool for analyzing complex biological data and predicting gene functions.