Ablation Hypothesis Quality Evaluation on Averaged BioLORD, GEARS, and CPA
1Semantic CoherenceAblateCell
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| AblateCellLLM Backend=GPT-4o-mini2026.04 | 1 | 0.92 | 0.867 | |
| AblateCellLLM Backend=Claude-Sonnet-4.52026.04 | 1 | 0.948 | 0.933 | |
| Heuristic-based2026.04 | 0.685 | 0.643 | 0.4 | |
| Random selection2026.04 | 0.428 | 0.357 | 0.133 |