Paper: Devkota, Shonai, Mao et al. 2026, Nature (Open Access) DOI / URL: 10.1038/s41586-026-10842-8 Accession(s): Zenodo 19546626 · figshare 33107414 · GitHub rohitsinghlab/raygun · AddGene article 28275322 (gated) Source / host: Zenodo (primary) · figshare · GitHub · Springer Nature · AddGene Disease relevance: n/a — generative protein-design methods paper (category: Protein sequence / Machine learning) Modality / assay: other (template-guided generative protein design) · Platform: ESM-2 (650M) PLM + Raygun autoencoder (701M params) Organism: multiple (UniRef50 training set; validated on eGFP/Aequorea victoria, mCherry/Discosoma, TurboID/E. coli-derived, human/mouse EGF) Samples: ~80,000 UniRef50 proteins (training); designed/validated candidates deposited as 19 AddGene plasmids Size: 1.4 GB (1519313410 bytes) Access / raw: public / OPEN-raw (AddGene deposit is gated/physical)
Download
- Zenodo (training + benchmarking FASTAs): https://zenodo.org/records/19546626
- figshare (results & figure-generation code/notebooks): https://doi.org/10.6084/m9.figshare.33107414
- GitHub (source code, pinned commit
cd3b3574708a71f6dc134c7719d287b5ab274186): https://github.com/rohitsinghlab/raygun - AddGene (19 deposited plasmids — physical/gated): https://www.addgene.org/browse/article/28275322/
- Supplements (Springer Nature): Supplementary Information, Reporting Summary, Peer Review File
- PMIDs: none assigned at crawl time
Data modalities & assays
Computational protein design. Raygun encodes each protein as a fixed-dimension probability distribution over ESM-2 embeddings, enabling single-shot miniaturization, magnification and modification of template proteins while preserving predicted structure/function. No sequencing/omics assay data; the "data" is training/benchmark sequence FASTAs, figure-reproducibility results/code, source code, and physically deposited validation plasmids.
Per-source breakdown
- Zenodo — 19546626 · access: public ·
1 file (
training_fastas.tar.gz, 941.7 MB): UniRef50 training splits (100K / 2.2M), SCOP structure-benchmarking set, and active/binding-site benchmark FASTAs. Primary deposited dataset (ref. 51; DOI 10.5281/zenodo.19546626). - figshare — 33107414 · access: public ·
1 file (
figure-reproducibility.tar.gz, 493.2 MB): per-figure results, plots, SVG/AI artwork, CSVs and.py/.ipynbcode that generate the paper's figures. (Cited in the GitHub README, not the Data-availability statement.) - GitHub — rohitsinghlab/raygun · access: public ·
1 file (repo archive @
cd3b357, 10.4 MB): Raygun package, pretrained-model loaders, training/sampling CLIs, example configs, notebooks, and template/benchmark FASTAs (incl. eGFP/mCherry). CC-BY-NC-SA 4.0. - Springer Nature supplements · access: public · 3 PDFs: Supplementary Methods & Notes, Reporting Summary, Peer Review File.
- AddGene — article 28275322 · access: restricted/physical · 19 deposited plasmids (11 pCAG-Raygun-TurboID, 4 pCAG-Raygun-eGFP, 4 pCAG-Raygun-mCherry). Sequences are served through an access-controlled SnapGene viewer (not machine-downloadable); physical DNA requires an MTA. See Gated access below. The corresponding designed protein sequences are available machine-readable via the GitHub and figshare bundles.
Files
| File | Size | SHA256 |
|---|---|---|
input/zenodo/training_fastas.tar.gz | 941.7 MB | c339d4e05493c443… |
input/figshare/figure-reproducibility.tar.gz | 493.2 MB | 94b58b850990b176… |
input/github/rohitsinghlab-raygun-cd3b357.tar.gz | 10.4 MB | 361293e1cb8f33a3… |
input/supplement/41586_2026_10842_MOESM1_ESM.pdf | 550.4 KB | fc75d5ea0a65fc55… |
input/supplement/41586_2026_10842_MOESM2_ESM.pdf | 1.1 MB | b9f6aee4a2dcab82… |
input/supplement/41586_2026_10842_MOESM3_ESM.pdf | 2.1 MB | 41fa6553fcf64b39… |
input/addgene/README_ADDGENE.md | 3.8 KB | 2d848cef8c4f3573… |
Index note: Raygun generative protein design (ESM-2 + autoencoder). Primary source Zenodo 19546626; multi-source (figshare + GitHub + Nature supplements). Paper PDF sha256
0fc3b6e4821ac1c0156d70ea11eb235067f743ea80a59e295570cf4b5de105d3.
Gated access
AddGene article 28275322 deposits the 19 experimentally validated, Raygun-designed
plasmids (fluorescent proteins + TurboID biotin ligases) as physical DNA. Per-plasmid
"Addgene-Verified" sequences are only viewable via AddGene's interactive SnapGene viewer
(GenBank downloads return HTTP 403/404 to scripted clients and require login/agreement),
so they are recorded as gated. Full plasmid list and request instructions are in
input/addgene/README_ADDGENE.md. The same designed
protein sequences are available machine-readable in input/github/ and input/figshare/.
Crawled by paper-to-datalake-agent v1.0 at 2026-08-04T05:16:04Z.
About this dataset
This Generative protein design (ESM-2 embeddings; Raygun autoencoder) dataset provides single-cell gene expression from human and mus musculus and aequorea victoria and discosoma and escherichia coli samples. 10x Genomics published it on July 29, 2026. The data were processed with Raygun 0.2.5, and the 0 output files total 1.4 GB.
Analyze it in Drylab
Open this dataset in Drylab and ask a question in plain language. The AI agent loads the files, runs the analysis in a ready sandbox and explains each result. A typical first pass:
- Load the filtered feature-barcode matrix (HDF5) and run quality control on counts, genes and mitochondrial reads per cell.
- Normalize, select variable genes, then cluster the cells and draw a UMAP.
- Annotate clusters with marker genes and compare the cell-type mix between conditions.