Benchmarking single-cell format conversion tools reveals interoperability gaps and introduces CrossCell for reliable
Zhi Wang1, Yan Zhang1, Zhengwei Wu1
1Department of Oral and Maxillofacial Surgery, Affiliated Hospital of Medical School, Institute of Stomatology, Nanjing Stomatological Hospital, Nanjing University, Jiangsu, China.
Background:
Single-cell RNA sequencing (scRNA-seq) analysis relies on two ecosystems-R (Seurat/Bioconductor) and Python (Scanpy/AnnData)-whose formats, RDS and H5AD, are not natively interoperable. Although conversion tools exist, no systematic benchmark has evaluated their reliability, performance, or data fidelity across diverse datasets and directions.
Results:
We present a systematic and comprehensive benchmark of five tools (Zellkonverter, anndataR, convert2anndata, easySCF, CrossCell) across seven dimensions: conversion robustness, runtime performance, memory efficiency, expression fidelity, metadata type integrity, Seurat V5 compatibility, and large-scale scalability. The benchmark covers 68 datasets ranging from 1,438 to 1,665,937 cells. We identify a substantial gap in H5AD-to-RDS conversion. Cross-language Foreign function interface boundaries introduce implicit type changes (e.g., Int32 to Float64) that can affect downstream data compliance. Existing tools already fail extensively on datasets below 74,000 cells, leaving large-scale conversion unaddressed. CrossCell-a Rust-based, runtime-independent toolkit introduced as part of this benchmark - achieves 100% success across all 55 standard datasets, delivers 2.5-4.9 × speedups and 37-61% memory savings over competing tools, and successfully converts 13 CELLxGENE datasets up to 1.67 million cells.
Conclusions:
This study exposes critical interoperability gaps and failure modes, providing the community with tool selection guidelines and best practices for cross-ecosystem data exchange. The fully containerized benchmark suite (68 datasets, 7 dimensions, one-click reproducibility) constitutes the first reproducible evaluation framework for this domain and can be directly applied to assess future tools. The runtime-independent design paradigm offers a viable path toward resolving cross-language data exchange bottlenecks.
