Upload README.md with huggingface_hub

README.md

@@ -19,6 +19,7 @@ Checkpoints of CELL-FM.
 | `condenseq/vae.bin` | Image VAE, 160x160, 3 down blocks, 4 latent channels |
 | `condenseq/vit_cls.bin` | ViT condensed/diffuse classifier, 2-channel 160x160 input |
 | `condenseq/reference_nucleus.npy` | The DAPI channel every CondenSeq generation is conditioned on: `(1, 160, 160)` float32 in [-1, 1]. CondenSeq protein index 12626, image 0 — the same one the offline seq2img runs used |
+| `condenseq/substitution_scan.npz` | The amino-acid substitution scan behind Fig. 5d (Δ AUC) and Supplementary Fig. 5c (Δ AAC): 100 held-out CondenSeq IDPs, every residue type replaced by each other type, 512 images per sequence. `amino_acids` (20,) in the figures' order R H K D E S T N Q G A V I L M C F Y W P, `idr_index` and `wt_sequence` (100,), `wt_auc` and `wt_aac` (100,), and `mut_auc` and `mut_aac` (20, 20, 100) indexed [from, to, IDP] — NaN on the diagonal and where the IDP has no `from` residue. The figures' matrices are the mean of `mut − wt` over the non-NaN IDPs; AUC and AAC are normalized, AAC being `compute_area_if_no_reentrant_minus_auc` |
 | `hpa/cellfm_seq2img.bin` | CELL-FM sequence-to-image generator for HPA, trained with test proteins held out; 256x256, 3-channel conditioning (includes the ESM-C 600M encoder) |
 | `hpa/cellfm_vs.bin` | CELL-FM virtual-staining generator for HPA, trained on all HPA proteins; 256x256, 3-channel conditioning (includes the ESM-C 600M encoder) |
 | `hpa/vae.bin` | Image VAE at 256x256, the one both `hpa/cellfm_seq2img.bin` and `hpa/cellfm_vs.bin` were trained against |