Learning pair-wise gene functional similarity by multiplex gene expression maps.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Extensive research has explored gene functional similarity with expression profiles and sequence data.
• Limited studies have investigated the link between gene functions and their expression locations in mammalian tissues.
• Supervised learning is challenging in functional genomics due to the lack of predefined target attributes for normal genes.

Purpose of the Study:

• To propose a supervised learning methodology for predicting pair-wise gene functional similarity.
• To leverage multiplex gene expression maps, which include spatial expression information, for this prediction.
• To explore the relationship between the similarity of gene expression maps and actual gene functions.

Main Methods:

• Developed a supervised learning approach using gene expression maps as input.
• Extracted features from expression maps, including wavelet features, raw expression values, and spatial neighborhood information (difference/average of voxels).
• Employed AdaBoost with a novel weak classifier to analyze large-scale gene expression datasets and predict functional similarities.

Main Results:

• Demonstrated a positive correlation between the similarity of gene expression maps and pairwise gene functional similarity.
• The developed model achieved a degree of accuracy in predicting gene functional similarities.
• Identified significant voxels and neighboring voxel pairs, visualizing their importance in mouse brain expression maps.

Conclusions:

• The proposed supervised learning method effectively predicts pair-wise gene functional similarity from spatial expression data.
• The findings highlight the utility of expression map similarity as a predictor of functional similarity.
• The methodology offers a valuable tool for predicting functions of unknown genes and has broader applications for analyzing large datasets lacking target attributes.