Upload _paper_results/cycle_70zz44_moe_endtoend_v2_summary.md with huggingface_hub
Browse files
_paper_results/cycle_70zz44_moe_endtoend_v2_summary.md
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# MoE end-to-end v2 (n=600 mixed: 200 T1 + 200 T2 + 200 T3)
|
| 2 |
+
|
| 3 |
+
- T1 trunk: /workspace/dnathinker/runs/exp_t1_ntv3_100m_rich_cond_corr_extra_v5full15m_fixed_20260501_101746/best.pt
|
| 4 |
+
- T2 ckpt: /workspace/dnathinker/runs/exp_t2_ntv3_dual_xa_FT100M_combined_b8_20260504_080327/best.pt (dual-XA Variant C, paper headline)
|
| 5 |
+
- T3 trunk: /workspace/dnathinker/runs/exp_t3_edit_tight_20260505/best.pt
|
| 6 |
+
|
| 7 |
+
## T1 enhancer generation (n=200)
|
| 8 |
+
|
| 9 |
+
- FID: 2.8697
|
| 10 |
+
- on_target_score: -1.9628
|
| 11 |
+
- off_target_score: -2.2795
|
| 12 |
+
- specificity: 0.3168
|
| 13 |
+
- argmax_accuracy: 0.1600
|
| 14 |
+
- diversity_edit: 0.5304
|
| 15 |
+
- emb_cosine_similarity: 0.7362
|
| 16 |
+
|
| 17 |
+
## T2 pair prediction (dual-XA Variant C, n=200)
|
| 18 |
+
- AUROC: 0.4427
|
| 19 |
+
|
| 20 |
+
## T3 enhancer editing (production-route BASE, n=200)
|
| 21 |
+
# T3 enhancer-editing motif scoring
|
| 22 |
+
|
| 23 |
+
**Predictions**: `/workspace/dnathinker/runs/eval_moe_endtoend_v2_n600_20260506_155742/predict_t3.jsonl`
|
| 24 |
+
**Gold**: `/workspace/dnathinker/data/full_enriched_v2_with_enh_scan/jsonl/test.enhancer_editing.jsonl`
|
| 25 |
+
**N**: 200, parse_rate=1.000
|
| 26 |
+
|
| 27 |
+
## Headline
|
| 28 |
+
|
| 29 |
+
| Metric | Value |
|
| 30 |
+
|---|---:|
|
| 31 |
+
| Mean predicted motif hits / seq | **0.650** |
|
| 32 |
+
| Mean gold motif hits / seq | 1.039 |
|
| 33 |
+
| Pred / Gold ratio | 0.626 |
|
| 34 |
+
| Target-TF motif hit rate | **0.110** (22/200) |
|
| 35 |
+
| Mean predicted GC | 0.530 (gold: 0.528) |
|
| 36 |
+
|
| 37 |
+
## Per-cell
|
| 38 |
+
|
| 39 |
+
| Cell | N | Pred motifs/seq | Gold motifs/seq | Pred GC | Gold GC |
|
| 40 |
+
|---|---:|---:|---:|---:|---:|
|
| 41 |
+
| Mic | 45 | 0.822 | 0.689 | 0.541 | 0.540 |
|
| 42 |
+
| End | 8 | 2.500 | 2.750 | 0.534 | 0.534 |
|
| 43 |
+
| Ex | 47 | 0.787 | 0.362 | 0.510 | 0.508 |
|
| 44 |
+
| OPC | 18 | 0.833 | 0.944 | 0.574 | 0.575 |
|
| 45 |
+
| Ast | 27 | 0.037 | 0.333 | 0.550 | 0.549 |
|
| 46 |
+
| In | 25 | 0.280 | 0.000 | 0.523 | 0.520 |
|
| 47 |
+
| Oli | 30 | 0.433 | 0.367 | 0.467 | 0.467 |
|