Measurement of Conditional Relatedness Between Genes Using Fully Convolutional Neural Network

Area of Science:

• Bioinformatics and Computational Biology
• Genomics and Systems Biology
• Machine Learning in Biology

Background:

• Traditional co-expression analysis for gene relatedness suffers from high false positive rates.
• Integrating prior-knowledge similarities is a viable strategy, but classical machine learning struggles with complex relationships.
• A powerful predictive model is needed to accurately map similarities to conditional gene relatedness.

Purpose of the Study:

• To develop a novel deep learning model for measuring conditional gene relatedness.
• To improve the accuracy of identifying gene-gene interactions and constructing biological networks.
• To overcome limitations of existing methods in handling complex relationships between gene similarities.

Main Methods:

• Proposed a fully convolutional neural network (FCNN) model, a deep learning approach.
• Utilized both co-expression and prior-knowledge similarities as input features.
• Employed grid-search 10-fold cross-validation for model evaluation.

Main Results:

• The FCNN model achieved higher accuracy in identifying gene-gene interactions across multiple benchmark datasets (COXPRESdb, KEGG, TRRUST, Xiao-Yong et al., GeneFriends, DIP).
• Demonstrated average accuracy improvements of 3.0%, 2.7%, 1.8%, and 7.6% on specific datasets.
• Successfully applied the FCNN model to construct cancer gene networks, outperforming other methods.

Conclusions:

• The FCNN model offers a significant advancement in measuring conditional gene relatedness.
• This deep learning approach provides more accurate gene-gene interaction detection and network construction.
• The FCNN model presents a powerful tool for bioinformatics research, with a publicly accessible website.