Copied


TabBioMed Benchmark Introduced for Biomedical Tabular Learning

Peter Zhang   Nov 20, 2026 08:42 0 Min Read


A new large-scale benchmark for biomedical tabular learning, TabBioMed, has been introduced to standardize evaluation across 94 curated public datasets. The initiative, detailed in a paper published on November 20, 2026, on arXiv (2611.23779), aims to address fragmented evaluation practices in biomedical machine learning. The benchmark spans diverse domains, including electronic health records, drug discovery, genomics, transcriptomics, proteomics, metabolomics, single-cell omics, and systems biology.

Standardized Framework and Dataset Scope

TabBioMed unifies datasets through a standardized preprocessing and evaluation framework, enabling reproducible comparisons for classification, regression, and multi-target tasks. The benchmark includes datasets with high dimensionality, large sample sizes, missing data, and class imbalances, reflecting real-world challenges in biomedical prediction.

Model Performance and Findings

The research benchmarks 23 models, including classical baselines, gradient-boosted trees, neural tabular architectures, and tabular foundation models. Foundation models demonstrated the strongest family-level predictive performance. However, the study highlights that tree-based models and tuned multilayer perceptrons remain competitive, particularly under constraints related to inference time or training costs. This underscores the context-dependent nature of optimal model selection in biomedical machine learning.

Open-Source Implementation

The TabBioMed benchmark is available as an open-source tool on GitHub (https://github.com/AI-sandbox/TabBioMed). It includes curated datasets, preprocessing code, baseline implementations, and results. Users can install the benchmark using Python 3.12 and access specific model groups, such as gradient-boosted trees or neural architectures, through modular installation commands. The repository also provides tools for downloading and verifying datasets to ensure reproducibility.

Implications for Biomedical Machine Learning

By offering a reproducible and comprehensive evaluation framework, TabBioMed facilitates practical model selection and supports future methods development in biomedical machine learning. Its release is expected to advance research and applications in fields such as drug discovery, genomics, and healthcare analytics.


Read More