Detection of Spatially Variable Genes: A Benchmark Study

In recent years, the field of OMICS sciences has undergone significant advancements due to the emergence of technologies capable of capturing biological data with higher precision. Among these, Spatial Transcriptomics (ST) has introduced the ability to retain spatial context in RNA expression data, offering new insights into tissue organization and function. A key aspect of ST analysis is the identification of Spatially Variable Genes (SVGs), whose expression levels vary across spatial locations and are critical for understanding complex biological processes. However, detecting SVGs remains a challenging task due to the diversity of computational methods and tools available, often implemented in different programming languages and based on distinct assumptions. This thesis presents a comprehensive benchmarking framework for SVG detection methods, evaluating them through public and simulated datasets and employing a reproducible Snakemake pipeline. By comparing statistical, graph-based, and machine learning approaches, this work aims to provide an objective foundation for method selection in ST analysis. The results offer valuable insights into the strengths and limitations of current techniques and contribute to enhancing the reliability and reproducibility of spatial transcriptomics studies.