pengxunduo/Geneformer-RNAquarium

⚠️ This is a derivative fork of Geneformer — not the official repository.

It accompanies the RNAquarium manuscript (Aniseia, Walteri et al.) and provides the Geneformer package with a small set of modifications used in that work. The figure-generation code is maintained separately and references this fork.

• Forked from: ctheodoris/Geneformer at the gc95M generation (default model GF-12L-95M-i4096), upstream commit 31bf641 (2025-06-16), prior to the upstream gc104M migration.
• What changed: see CHANGES_vs_upstream.md. Edits are confined to 6 files in geneformer/ — cross-model weight transfer, class subsampling/weighting, multi-GPU multitask training, and enhanced confusion-matrix plotting.
• Model weights are not included here. Download pretrained Geneformer weights from upstream: https://huggingface.co/ctheodoris/Geneformer. RNAquarium-specific trained weights, if shared, are hosted separately (see the manuscript).
• License & attribution: Apache-2.0, inherited from upstream. See LICENSE and NOTICE.

---