On-domain cell identity prediction on LivST1
77Macro F1CellRefine → FF
Evaluation Results
| Method | Links | |
|---|---|---|
| CellRefine → FFPost-training=CellRefine, Fine-tuning=Full2026.05 | 77 | |
| CellRefine LoRA → LoRAPost-training=CellRefine, Fine-tuning=LoRA2026.05 | 71 | |
| MLM → FFPost-training=MLM, Fine-tuning=Full2026.05 | 68 | |
| FFStrategy=Full Fine-Tuning2026.05 | 64 | |
| MLM LoRA → LoRAPost-training=MLM, Fine-tuning=LoRA2026.05 | 51 | |
| LoRAStrategy=Parameter-efficient fine-tuning2026.05 | 50 | |
| MLM LL → LLPost-training=MLM, Fine-tuning=Last-Layer2026.05 | 42 | |
| CellRefine LL → LLPost-training=CellRefine, Fine-tuning=Last-Layer2026.05 | 40 | |
| LLStrategy=Last-Layer Fine-Tuning2026.05 | 37 | |
| LPStrategy=Linear Probing2026.05 | 13 |