Deciphering Cell Lineage Gene Regulatory Network via MTGRN

Rui Peng, Wei Wu, Jinzhuo Wang

Keywords: Gene regulatory network, Time series, In silico perturbation

Abstract: Gene regulatory network (GRN) inference is crucial for cell fate decision, as it outlines the regulations between genes, which direct cell differentiation. Although there have been some work to infer cell lineage GRN, they fail to capture the continuous nature of the differentiation process as they group cells by cell type or cluster and infer GRN in a discrete manner. In this paper, we hypothesize GRN can forecast future gene expression based on history information and transform the inference process into a multivariate time series forecasting problem, linking cells at different time to learn temporal dynamics and inferring GRN in a continuous process. We introduce MTGRN, a transformer-based model that only takes single cell data as input to infer the cell lineage GRN by forecasting gene expression. MTGRN consists of temporal blocks and spatial blocks, effectively captures the connections between cells along their developmental trajectories and leverages prior knowledge to elucidate regulatory interactions among genes. It significantly outperforms six other methods across five datasets, demonstrating superior performance even compared to multimodal approaches. Based on the inferred GRN, MTGRN pinpoints three crucial genes associated with the development of mouse embryonic stem cells and depicts the activity changes of these genes during cellular differentiation. Beyond this, MTGRN is capable of conducting perturbation experiments on key genes and accurately modeling the change of cell identity following the knockout of the Gata1 in mouse hematopoietic stem cells.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 5401