Machine learning-optimized targeted detection of alternative splicing.

Area of Science:

• Molecular Biology
• Genomics
• Bioinformatics

Background:

• RNA-sequencing (RNA-seq) is a standard tool for transcriptome analysis.
• RNA-seq faces challenges in comprehensively detecting and quantifying alternative splicing due to inherent biases.
• Accurate analysis of alternative splicing is crucial for understanding gene expression and function.

Purpose of the Study:

• To develop an efficient targeted RNA-sequencing method for improved alternative splicing analysis.
• To enhance the detection and quantification of splicing-informative junction-spanning reads.
• To address the limitations of standard RNA-seq in alternative splicing studies.

Main Methods:

• Introduction of Local Splicing Variation sequencing (LSV-seq), a targeted RNA-seq technique.
• Utilizing multiplexed reverse transcription with primers anchored near splicing events.
• Employing Optimal Prime, a machine learning algorithm for primer design.
• Applying deep learning splicing code predictions to target low-coverage events.

Main Results:

• LSV-seq demonstrates high on-target capture rates and concordance with standard RNA-seq.
• The method achieves significant improvements in sensitivity with substantially lower sequencing depth.
• LSV-seq enabled the discovery of hundreds of novel tissue-specific splicing events in GTEx data.
• High-throughput quantification of splicing events with exceptional sensitivity was achieved.

Conclusions:

• LSV-seq is an efficient and sensitive method for targeted alternative splicing analysis.
• The technique overcomes limitations of standard RNA-seq, offering deeper insights into splicing variations.
• LSV-seq facilitates high-throughput discovery of tissue-specific splicing events.