Omics & gene networks

Decoding the Genetic Regulation landscape of chicken single cells using foundational models

---

Abstract

Foundational AI models have revolutionized biological research, yet their application to avian species like the chicken, a critical model for development and agriculture, remains limited. To address this, we leveraged the Chicken Cell Atlas (ChCA), comprising over six million single cells from 56 tissues, to develop a chicken-specific single-cell foundation model based on the xtrimo-cell architecture and fine-tuned the Borzoi DNA model using a pseudobulk expression matrix. Our single-cell model converged stably, outperforming baseline benchmarks in cell type annotation and generating robust gene embeddings that recapitulated canonical pathways and TF motif enrichments. Furthermore, the fine-tuned Borzoi model achieved high predictive accuracy (Pearson r≈0.78), enabling the corroboration of single-cell eQTLs mapped via deconvolution of Advanced Intercross Line (AIL) data. Collectively, these results demonstrate the feasibility of chicken-specific foundation models, providing a powerful framework to delineate the cell-type–specific regulatory mechanisms underlying complex traits in non-mammalian models.

Keywords: 2026

How to Cite:

Guo, Y., Hu, X., Li, C., Wang, Y., Yang, Q. & Zhai, C., (2026) “Decoding the Genetic Regulation landscape of chicken single cells using foundational models”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285367. doi: https://doi.org/10.31274/wcgalp.23659

Rights: 1