Multisource omic alignment and biological feature discovery with Performer encoder and triplet networks

Area of Science:

• Genomics
• Bioinformatics
• Developmental Biology

Background:

• Single-cell sequencing advances molecular understanding but integrating multiomic data across species for key factor discovery remains challenging.
• Comparative studies in livestock, like pigs and cattle, require robust methods for cross-species analysis.

Purpose of the Study:

• To develop a deep learning model, AlignCell, for integrating multisource omic data across platforms, omic types, and species.
• To facilitate the discovery of conserved and species-specific genes in cross-species comparative studies, particularly in livestock.

Main Methods:

• Developed AlignCell, a deep learning model for integrating diverse omic datasets.
• Applied AlignCell to integrate female gonad data across human, mouse, pig, and cattle, including new bovine single-cell data.
• Benchmarked AlignCell against existing tools for performance evaluation.

Main Results:

• AlignCell demonstrated strong performance in integrating multiomic data and identifying key biological factors.
• Integration of female gonad data revealed the conserved gene CCT2 in primordial germ cells (PGCs) across species.
• Identified novel species-specific genes, PRICKLE4 in pigs and CTSV in cattle PGCs.

Conclusions:

• AlignCell is an effective tool for deep learning-based integration of multisource omic data across species.
• The study identified novel conserved and species-specific genes in PGCs, offering insights into reproductive and developmental biology.
• AlignCell facilitates cross-species comparative genomics, particularly for livestock research.