DOLPHIN advances single-cell transcriptomics beyond gene level by leveraging exon and junction reads

Area of Science:

• Genomics
• Computational Biology
• Molecular Biology

Background:

• Single-cell sequencing offers high-resolution insights into cellular heterogeneity.
• Current methods often underutilize exon-level and junction read data, limiting analysis depth.
• Critical transcriptomic details are missed by conventional gene-count approaches.

Purpose of the Study:

• To introduce DOLPHIN, a novel deep learning method for enhanced single-cell RNA sequencing analysis.
• To leverage exon-level and junction read data for improved cell representation and downstream applications.
• To overcome limitations of traditional gene-count methods in capturing cellular dynamics.

Main Methods:

• Developed DOLPHIN, a deep learning framework integrating exon and junction read data.
• Represented genes as graph structures for variational graph autoencoder processing.
• Utilized graph autoencoders to generate improved cell embeddings.

Main Results:

• DOLPHIN demonstrated superior performance in cell clustering and biomarker discovery.
• The method showed enhanced accuracy in alternative splicing detection.
• DOLPHIN identified subtle, exon-level transcriptomic differences often masked by gene-level analyses.

Conclusions:

• DOLPHIN significantly advances single-cell transcriptomic analysis by incorporating detailed exon and junction information.
• The method provides a more nuanced understanding of cellular dynamics and heterogeneity.
• DOLPHIN offers new avenues for disease mechanism research and therapeutic target identification.