scikit-bio roadmap

This is a dynamic document tracking current development directions. It may be used for inspiration and encouragement and for coordination of efforts.

Type annotations

We are systematically adding type annotations to the scikit-bio codebase to improve code readability, IDE autocompletion, and bug detection in both development and downstream applications. Recent progress includes type annotations for distance statistics, alignment modules, and other core functionality shipped in v0.7.0. New code development incorporates type annotations by default, while existing modules are being updated incrementally.

Interoperability

We are enhancing scikit-bio’s ability to work seamlessly with diverse data formats and computing frameworks. A new dispatch system handles multiple table formats including arrays, Pandas and Polars dataframes, BIOM tables, and AnnData objects, allowing functions to operate on any supported format while preserving input types or enabling user-specified output formats. This table dispatch system shipped in v0.7.0. Additionally, we are adopting the Python array API standard to enable compatibility with GPU-accelerated libraries. The integration of these complementary systems will significantly improve scikit-bio’s interoperability across the scientific Python ecosystem.

GPU support

Modern omics datasets often exceed memory limitations and benefit from hardware acceleration. To address this, we are adopting the Python array API standard, which enables seamless interoperability with GPU-accelerated libraries like CuPy and JAX, as well as distributed computing frameworks like Dask. Initial implementation has focused on compositional transformation algorithms, with successful testing on PyTorch GPU objects. This foundation will be extended to additional computational bottlenecks including UniFrac, PERMANOVA, mantel, and permdisp, enabling significant performance improvements for large-scale analyses.

Support for free-threaded Python

CPython 3.13 introduces a free-threaded build that removes the Global Interpreter Lock, enabling true multi-threading for CPU-bound tasks. This presents significant opportunities for scikit-bio’s computationally intensive operations. However, adoption depends on ecosystem readiness - key dependencies including statsmodels and h5py do not yet provide free-threaded builds. We will monitor dependency support and evaluate implementation once the required packages become available, with particular focus on algorithms that would benefit most from parallel execution.

Benchmarking

Performance monitoring is critical for preventing regression in computationally intensive bioinformatics workflows. While scikit-bio previously had lightweight benchmarks using asv, these were not maintained and became outdated. We have established a dedicated benchmarking repository (scikit-bio/scikit-bio-benchmarks) to systematically track performance across releases, starting with version 0.6.1 and continuing for all future releases. The current framework includes basic benchmarks with plans to expand coverage to key algorithms and data structures. This infrastructure will help ensure that performance improvements and potential regressions are identified early in the development process.