Understanding how genes work together inside a cell is one of the central challenges in biology. Genes rarely act alone, instead they form complex gene regulatory networks that control everything from cell growth to response to disease. Mapping these networks helps scientists better understand how cells function and how diseases develop.

With the rise of single-cell RNA sequencing, researchers can now measure gene activity in individual cells at an unprecedented level of detail. This has opened the door to studying how gene networks operate in different cell types and conditions. However, turning this data into accurate maps of gene interactions remains difficult, especially when trying to incorporate experimental perturbations where genes are intentionally altered.

Researchers from Michigan Medicine have developed a new method called PSGRN to improve how gene regulatory networks are inferred from single-cell data. This approach combines both observational data and perturbation data, allowing it to capture more meaningful biological relationships between genes.

One of the key innovations in PSGRN is its use of a self-training framework. The model generates synthetic reference data, known as gold standards, to guide its learning process. This helps the system better distinguish true gene interactions from noise and improves overall accuracy.

When tested across multiple datasets, PSGRN consistently outperformed existing methods. It showed higher precision and recall, meaning it was better at correctly identifying real gene interactions while minimizing errors. These improvements are important because even small inaccuracies in network models can lead to misleading biological conclusions.

Another advantage of PSGRN is its scalability. As single-cell datasets continue to grow in size and complexity, methods that can efficiently handle large amounts of data are essential. PSGRN provides a flexible and robust framework that can be applied across different biological systems and experimental designs.

Overall, this work highlights how combining advanced computational methods with single-cell RNA sequencing can lead to more accurate models of gene regulation. These insights can ultimately support better understanding of disease mechanisms and guide the development of new therapies.

Availability – The codes are released via Zenodo (https://doi.org/10.5281/zenodo.17996079) and can be accessed through GitHub (https://github.com/GuanLab/PSGRN).