Spider: a flexible and unified framework for simulating spatial transcriptomics data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Spatial transcriptomics (ST) technologies offer insights into cellular heterogeneity by combining gene expression and spatial location data.
• Existing "gold standard" datasets for benchmarking ST analysis tools lack diversity and accuracy, limiting tool evaluation.
• There is a need for robust and flexible methods to simulate ST data for reliable tool development and validation.

Purpose of the Study:

• To introduce Spider, a novel framework for simulating spatial transcriptomics data.
• To enhance the realism, diversity, and flexibility of simulated ST data compared to existing methods.
• To provide a tool that facilitates the benchmarking and evaluation of ST analysis tools.

Main Methods:

• Spider simulates ST data by characterizing spatial patterns using cell type proportions and a transition matrix between adjacent cells.
• The framework allows for interactive customization of the spatial domain, including zone segmentation and integration of histology imaging.
• No real ST data is required as a reference for data simulation.

Main Results:

• Spider generates more realistic and diverse simulated ST data with enhanced modeling flexibility.
• Benchmark analyses show Spider preserves spatial characteristics of real ST data better than other simulation tools.
• Spider effectively facilitates the evaluation of downstream ST analysis methods.

Conclusions:

• Spider offers a flexible and comprehensive solution for simulating ST data, addressing limitations in current benchmarking datasets.
• The framework improves the reliability and fairness of evaluating ST analysis tools.
• Spider is publicly available, promoting reproducible research and further development in the field.