arXiv：cs.LG· Michael Bohl, Alexander Theus, David Wissel, Valentina Boeva·· 4 小时前AI 评分39
GeneICL：面向批量转录组学的表格基础模型
GeneICL: A Tabular Foundation Model for Bulk Transcriptomics
AI 导读
研究者推出 GeneICL，一个 4.2M 参数的表格基础模型，通过结合实测批量表达谱构建的半合成预训练先验与参数高效循环架构，实现转录组学感知的预训练。在覆盖分类、回归和生存预测的 80 项临床结果预测任务上，GeneICL 在评测的基础模型和调优基线中取得最佳整体排名，参数量最多减少 387 倍，推理无需梯度更新，笔记本 CPU 上数秒内即可完成预测。
正文
Abstract:Gene expression is widely measured in biomedicine, yet clinical outcome prediction remains challenging due to high dimensionality, strong feature correlations, and limited labeled data. Large self-supervised transcriptomic foundation models often fail to outperform simple supervised baselines. Tabular foundation models offer an alternative through in-context learning, but are typically pretrained on generic synthetic data rather than transcriptomic structure. We ask whether transcriptomics-aware pretraining, rather than scale, is the missing ingredient. Towards this end, we introduce GeneICL, a 4.2M-parameter tabular foundation model combining a semi-synthetic pretraining prior built from measured bulk expression profiles with a parameter-efficient recurrent architecture. We further enable right-censored survival prediction via a training-free reduction to regression using Cox partial-likelihood residuals. We evaluate GeneICL on 80 clinical outcome-prediction tasks spanning classification, regression, and survival. Tabular foundation models consistently outperform self-supervised transcriptomic models, while GeneICL achieves the best overall rank among evaluated foundation models and tuned baselines. GeneICL does so with up to 387$\times$ fewer parameters, no gradient updates at inference, and predictions within seconds on a laptop CPU.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.08694 [cs.LG] |
| (or arXiv:2610.08694v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.08694 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Alexander Theus [view email]
[v1]
Tue, 6 Oct 2026 17:10:37 UTC (1,183 KB)
来源：arXiv：cs.LG · arxiv.org