eheiss/scbFM_models

scbFM model checkpoints

Pretrained and pre-adapted model checkpoints for scbFM, accompanying Single-Cell Foundation Models for Bulk Transcriptomics: Evaluation and Adaptation.

Download

Download the scbFM checkpoints into the default output directory with:

hf download eheiss/scbFM_models --local-dir "$SCBFM_ROOT_DIR/output"

The published scGPT and BulkFormer checkpoints are not redistributed here. Download them from their original sources linked below.

File structure

$SCBFM_ROOT_DIR/
|-- output/
|   |-- pretrain_sc/
|   |   `-- pretrain_sc.pth
|   |-- pretrain_bulk/
|   |   `-- pretrain_bulk.pth
|   |-- preadapt_sc/
|   |   `-- preadapt_sc.pth
|   |-- preadapt_bulk/
|   |   `-- preadapt_bulk.pth
|   `-- scgpt_preadapt/
|       |-- args.json                       # from published scGPT
|       |-- vocab.json                      # from published scGPT
|       `-- last_model.pt
`-- other/
    `-- models/
        |-- args.json                       # published scGPT args.json
        |-- vocab.json                      # published scGPT vocab.json
        |-- best_model.pt                   # published scGPT best_model.pt
        `-- BulkFormer_147M.pt              # published BulkFormer checkpoint

The scGPT configuration and vocabulary are also required beside scgpt_preadapt/last_model.pt. Copy the downloaded args.json and vocab.json into that directory without renaming them.

scbFM checkpoints

File	Description	Download
pretrain_sc/pretrain_sc.pth	Controlled model pretrained on scRNA-seq	Available in this repository
pretrain_bulk/pretrain_bulk.pth	Controlled model pretrained on bulkRNA-seq	Available in this repository
preadapt_sc/preadapt_sc.pth	scRNA-seq-pretrained controlled model after bulkRNA-seq pre-adaptation	Available in this repository
preadapt_bulk/preadapt_bulk.pth	bulkRNA-seq-pretrained controlled model after additional bulkRNA-seq training	Available in this repository
scgpt_preadapt/last_model.pt	Published human scGPT model after bulkRNA-seq pre-adaptation	Available in this repository

Published model dependencies

Model	Required files	Original download
scGPT whole-human	args.json, vocab.json, and best_model.pt	Official scGPT download
BulkFormer-147M	BulkFormer_147M.pt	Official BulkFormer download