scPRINT: pre-training on 50 million cells allows robust gene network predictions.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Cellular processes are driven by complex macromolecular interactions.
• Inferring these gene networks is a significant challenge in cellular biology.

Purpose of the Study:

• To introduce scPRINT, a large cell model for gene network inference.
• To leverage foundation models for enhanced interpretability and usability in uncovering cellular biology.

Main Methods:

• Pre-training scPRINT on over 50 million cells from the cellxgene database.
• Utilizing innovative pretraining tasks and a novel model architecture.
• Employing large transformer models for biological data analysis.

Main Results:

• scPRINT demonstrates superior performance in gene network inference compared to state-of-the-art methods.
• Achieved competitive zero-shot abilities in denoising, batch effect correction, and cell label prediction.
• Highlighted connections between ion exchange, senescence, and chronic inflammation in benign prostatic hyperplasia.

Conclusions:

• scPRINT advances the capability of large transformer models for biological network inference.
• The model offers improved interpretability and usability for complex cellular biology research.
• scPRINT provides novel insights into disease mechanisms, such as in benign prostatic hyperplasia.