SpliceVec: Distributed feature representations for splice junction prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Accurate identification of splice junctions is crucial for understanding gene structure and alternative splicing.
• Current RNA-sequencing methods for splice junction identification are prone to errors due to read mapping inaccuracies.
• Existing computational models rely on manual feature extraction, which may not capture all relevant splicing signals.

Purpose of the Study:

• To introduce SpliceVec, a novel computational method for splice junction identification using distributed feature representation.
• To overcome the limitations of explicit and potentially biased feature extraction in existing models.
• To improve the accuracy and efficiency of splice junction detection for genomic analysis.

Main Methods:

• Developed SpliceVec, a method employing distributed representations inspired by natural language processing models.
• Utilized a multilayer perceptron for splice junction classification based on learned SpliceVec features.
• Investigated the optimal sequence context for feature extraction, comparing intronic sequences with flanking regions.

Main Results:

• SpliceVec effectively distinguishes true splice junctions from false positives.
• Including the entire intronic sequence provides better performance than using only flanking regions.
• The model demonstrates invariance to canonical and non-canonical splice junctions.
• SpliceVec maintains consistent performance on reduced and class-imbalanced datasets.

Conclusions:

• SpliceVec offers a robust and efficient computational approach for splice junction identification.
• The method enhances the analysis of gene structure and alternative splicing.
• SpliceVec provides a valuable tool for genomic research, adaptable to user-defined data.