Identifying prognosis-associated spatial patterns by integrating bulk RNA-seq and spatial transcriptomic data

Area of Science:

• Computational biology
• Cancer research
• Bioinformatics

Background:

• Spatial transcriptomics (ST) reveals intra-tumor heterogeneity.
• Bulk RNA sequencing (RNA-seq) data contains valuable clinical phenotype information.
• Linking spatial features with survival data is crucial for cancer prognosis.

Purpose of the Study:

• To develop a deep transfer learning framework, stSurvTrans, for identifying prognosis-associated spatial patterns.
• To harmonize ST and bulk RNA-seq data.
• To transfer clinical survival information from bulk samples to ST data.

Main Methods:

• Conditional variational autoencoder (CVAE) for data harmonization.
• Weibull module for survival information transfer.
• Benchmarking on simulated datasets and hepatocellular carcinoma (HCC) ST data.

Main Results:

• stSurvTrans demonstrated accuracy and superiority on simulated datasets.
• The framework successfully linked spatial patterns with survival information.
• Identified bile duct tumor thrombus in HCC ST data as a spatial structure associated with worse prognosis.

Conclusions:

• stSurvTrans is an effective deep transfer learning framework for integrating ST and bulk RNA-seq data.
• It enables the identification of spatial patterns linked to patient survival.
• This approach has potential for improving cancer prognosis and understanding tumor biology.