Foundation embeddings beat published GNN+KG synthetic-lethality methods on SynLethDB CV1/CV2/CV3 with 31× DepMap CRISPR enrichment
收藏Zenodo2026-05-09 更新2026-05-26 收录
下载链接：
https://zenodo.org/doi/10.5281/zenodo.20075610
下载链接
链接失效反馈官方服务：
资源简介：
A general-purpose text embedding (GenePT) plus a 5-line sklearn MLPClassifier head, fed simple [concat; abs-diff; Hadamard] pair encoding, outperforms every published graph-neural-network plus knowledge-graph synthetic-lethality prediction method on the standard SynLethDB CV1/CV2/CV3 benchmark (Wang 2024 protocol). Independent DepMap CRISPR co-essentiality validation: 13.0% strong/moderate co-essential vs 0.4% random baseline = 31.3× enrichment. Includes intrinsic-dimensionality finding (GenePT effective dim ≈ 256) and multi-foundation stacking ablation.
提供机构：
Zenodo创建时间：
2026-05-07
