Add six MARRI checkpoints and model card
Browse files- README.md +63 -0
- dyn_ft-rnafm_seed42/best_rna_physics_model.pth +3 -0
- dyn_ft-rnafm_seed43/best_rna_physics_model.pth +3 -0
- dyn_ft-rnafm_seed44/best_rna_physics_model.pth +3 -0
- generalization_interaction_locus_seed42/best_rna_physics_model.pth +3 -0
- generalization_interaction_pair_seed42/best_rna_physics_model.pth +3 -0
- generalization_transcript_seed42/best_rna_physics_model.pth +3 -0
README.md
CHANGED
|
@@ -1,3 +1,66 @@
|
|
| 1 |
---
|
| 2 |
license: cc-by-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: cc-by-4.0
|
| 3 |
+
tags:
|
| 4 |
+
- rna
|
| 5 |
+
- rna-rna-interaction
|
| 6 |
+
- pytorch
|
| 7 |
---
|
| 8 |
+
|
| 9 |
+
# MARRI model checkpoints
|
| 10 |
+
|
| 11 |
+
Trained checkpoints for MARRI (mechanism-aware multimodal framework for RNA--RNA
|
| 12 |
+
interaction prediction). All checkpoints share the same architecture (RNA-FM +
|
| 13 |
+
RNet-2D dual-tower encoder, opening branch, hybridization-gain MLP, axial
|
| 14 |
+
cross-attention classifier) and the same backbone configuration -- RNA-FM
|
| 15 |
+
fine-tuned, RNet-2D frozen, epoch-wise dynamic negative resampling -- differing
|
| 16 |
+
only in random seed and/or training data split.
|
| 17 |
+
|
| 18 |
+
Code: [**https://github.com/Gardeina/MARRI**](https://github.com/Gardeina/MARRI)
|
| 19 |
+
Interpretability/attention export for the headline model: https://huggingface.co/datasets/Xu-AI4Science/MARRI-interpretability
|
| 20 |
+
|
| 21 |
+
Each `.pth` file is a raw `state_dict()` for the interaction model (not a full
|
| 22 |
+
training checkpoint with optimizer state), loadable via
|
| 23 |
+
`model.load_state_dict(torch.load(path, map_location=device))`.
|
| 24 |
+
|
| 25 |
+
## Main-matrix checkpoints (N-clean dataset, random 8:1:1 split)
|
| 26 |
+
|
| 27 |
+
| File | Seed | Test ROC-AUC | Best epoch | Notes |
|
| 28 |
+
|---|---|---|---|---|
|
| 29 |
+
| `dyn_ft-rnafm_seed42/best_rna_physics_model.pth` | 42 | 0.7530 | 23 | **Headline model**, reported as the main MARRI result throughout the paper |
|
| 30 |
+
| `dyn_ft-rnafm_seed43/best_rna_physics_model.pth` | 43 | 0.7409 | 15 | Seed-repeat run, same recipe |
|
| 31 |
+
| `dyn_ft-rnafm_seed44/best_rna_physics_model.pth` | 44 | 0.7342 | 18 | Seed-repeat run, same recipe |
|
| 32 |
+
|
| 33 |
+
Mean +/- SD over these three seeds: ROC-AUC 0.7427 +/- 0.0095 (Table 1 in the paper).
|
| 34 |
+
|
| 35 |
+
## Leakage-resistant held-out generalization checkpoints (seed 42, own data split each)
|
| 36 |
+
|
| 37 |
+
Same recipe, each retrained from scratch on its own leakage-resistant held-out
|
| 38 |
+
split (a separate positive/negative pool per split, not the main N-clean
|
| 39 |
+
dataset -- see Methods/Supplementary for split construction).
|
| 40 |
+
|
| 41 |
+
| File | Held-out split | Test ROC-AUC | Best epoch |
|
| 42 |
+
|---|---|---|---|
|
| 43 |
+
| `generalization_interaction_pair_seed42/best_rna_physics_model.pth` | Unseen RNA--RNA pair | 0.7422 | 14 |
|
| 44 |
+
| `generalization_transcript_seed42/best_rna_physics_model.pth` | Unseen RNA/gene (most stringent) | 0.6630 | 13 |
|
| 45 |
+
| `generalization_interaction_locus_seed42/best_rna_physics_model.pth` | Unseen local interaction locus | 0.7502 | 19 |
|
| 46 |
+
|
| 47 |
+
These three correspond to Table 4 ("Generalization under leakage-resistant
|
| 48 |
+
held-out splits") in the paper.
|
| 49 |
+
|
| 50 |
+
## Citation
|
| 51 |
+
|
| 52 |
+
Publication metadata will be updated when available.
|
| 53 |
+
|
| 54 |
+
```bibtex
|
| 55 |
+
@article{MARRI_TBD,
|
| 56 |
+
title = {MARRI: A Mechanism-Aware Multimodal Learning Framework for RNA--RNA Interaction Prediction},
|
| 57 |
+
author = {AUTHOR_LIST_TBD},
|
| 58 |
+
journal = {JOURNAL_TBD},
|
| 59 |
+
year = {YEAR_TBD},
|
| 60 |
+
doi = {DOI_TBD}
|
| 61 |
+
}
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
## Contact
|
| 65 |
+
|
| 66 |
+
For questions about MARRI, please contact **2300393032@szu.edu.cn**.
|
dyn_ft-rnafm_seed42/best_rna_physics_model.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c49bd144427a8a1fa9fe519c36a3d8a4172ae114c03ef394af4905231e16b57c
|
| 3 |
+
size 490677043
|
dyn_ft-rnafm_seed43/best_rna_physics_model.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be8b2763e103828e942119f8caf150646f8cfcc89caca8411fc6a48738daab2f
|
| 3 |
+
size 490677043
|
dyn_ft-rnafm_seed44/best_rna_physics_model.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b525bc1b9f47f317b2625673f0b81ba0a524b1c35ef716d94886d4a24e3c7961
|
| 3 |
+
size 490677043
|
generalization_interaction_locus_seed42/best_rna_physics_model.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c7a741e3b3324a438bc5e6229265ce20b67ffaac05fc65db81ab1b41adfd1ba2
|
| 3 |
+
size 490677043
|
generalization_interaction_pair_seed42/best_rna_physics_model.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d0b67bd42183ef26de58577aa0c6192eadd19a738562e73d77be7a446babdfee
|
| 3 |
+
size 490677043
|
generalization_transcript_seed42/best_rna_physics_model.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:df18902a8c33558ca3fc900c3e073c56bc0fe9c9d9e9e824cbe31071d66a48d0
|
| 3 |
+
size 490677043
|