GeneICL: A Tabular Foundation Model for Bulk Transcriptomics
GeneICL, a 4.2M-parameter tabular model, leads 80 transcriptomic outcome tasks without inference-time training.
Researchers introduce GeneICL, a 4.2 million-parameter tabular foundation model for bulk transcriptomics. It pairs a semi-synthetic prior built from measured expression profiles with a parameter-efficient recurrent architecture, and reduces right-censored survival to regression using Cox partial-likelihood residuals. On 80 clinical outcome tasks, tabular foundation models beat self-supervised transcriptomic models, and GeneICL ranked best while using up to 387 times fewer parameters and predicting in seconds on a laptop CPU.
- 4.2M-parameter model pretrained on measured bulk expression structure
- Covers classification, regression, and right-censored survival prediction
- Best overall rank on 80 clinical outcome tasks
- Up to 387× fewer parameters and no inference-time gradient updates
- AI models
- GeneICL
From the original article
1 min read at arxiv.orgGene 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.
By Michael Bohl · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.