# ML-Guided Directed Evolution of the NucB Nuclease: 55,760-Variant Activity Landscape and NGS Counts (Google DeepMind)

## Facts

- Provider: Thomas, Belanger, Xu et al. (Google DeepMind and Google Research, with collaborators)
- License: CC-BY-4.0 (https://github.com/google-deepmind/nuclease_design)
- Stored at: Google Cloud Storage, gs://nuclease_design (us)
- Access: public, readable without an account
- DOI: https://doi.org/10.1101/2024.03.21.585615
- Category: Proteomics
- Source: https://doi.org/10.1101/2024.03.21.585615
- Page: https://scigantic.com/archive/97111039-569b-5e03-8084-033a5abc8b74
- Keywords: Proteomics, Google Cloud Storage

## About

Data from Thomas, Belanger et al., 'Engineering highly active and diverse nuclease enzymes by combining machine learning and ultra-high-throughput screening' (bioRxiv 2024), in which ML models (including a VAE and a CNN) proposed NucB nuclease variants that were screened in four rounds of ultra-high-throughput sorting. The bucket holds raw_count_data/ (NGS counts for pre-sort and post-sort populations, rounds G1 to G4), processed_fiducial_data/ (enrichment factors for synonymous DNA sequences of fiducial variants, used to build null distributions), processed_data/ (per amino acid sequence enrichment factors and p-values for g1.csv to g4.csv, and landscape.csv, which merges all rounds into multi-class catalytic activity labels for about 56,000 distinct sequences; the README describes the estimated activity of 55,760 variants), plate_data/ (low-throughput purified protein assays confirming hits), library_designs/ (which design method proposed each sequence), alignments/ (the multiple sequence alignment used for the VAE), cnn_models/ and analysis/. Load CSVs with pandas; the nuclease_design package and Colab notebooks reproduce the figures. DeepMind licenses all non-software materials CC BY 4.0; cite Thomas, Belanger et al. 2024.

## Load the data

```python
import gcsfs
fs = gcsfs.GCSFileSystem(token='anon')
fs.ls('nuclease_design/', detail=False)[:10]
```

Lists the files with anonymous access. Runs on any machine, with no cloud account.

## Data structure

Example file: `alignments/F1BV52_BACLI_e3_n8_mFirst.a2m`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/97111039-569b-5e03-8084-033a5abc8b74
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/97111039-569b-5e03-8084-033a5abc8b74
- Over MCP: `get_schema_card` with id `97111039-569b-5e03-8084-033a5abc8b74` at https://api.scigantic.com/mcp

## Related topics

- [Proteomics datasets](https://scigantic.com/topics/proteomics)
