DeepES

DeepES is a deep learning-based framework for enzyme screening. DeepES can identify orphan enzyme candidate protiens by focusing on biosynthetic gene clusters and KEGG RClass. DeepES uses protein sequences as inputs and evaluate whether the input genes contain biosynthetic gene clusters of interest.

Requirements

• python 3.9.19 (with following packages)
  • numpy 1.26.4
  • pandas 2.2.1
  • pytorch 1.13.0
  • biopython 1.83
  • fair-esm 2.0.0

By using environment.yml, you can build an anaconda environment exactly the same as this research.

conda env create -f environment.yml
conda activate deepes

Pretrained parameters

The model weights are available at Zenodo. Please download files and unzip.

Program Usage

Embed input protein sequences

python embed.py \
    --input_dir data \
    --output_dir output

• --input_dir: Please specify the path of your directory where the input data is located.
• --output_dir: Please specify the path to save outputs.
• --reuse: Whether to use the results of previous runs. default:True
• --batch_size: Batch size in embedding protein sequences. default:1
• --cuda: Whether to use a GPU. default:False
• --cpu_num: Number of threads. default:1

Calculate probabilities

python predict.py \
    --input_dir data \
    --output_dir output \
    --model_dir model \
    --rclass_list RC01053 RC00004,RC00014 RC01923

• --input_dir: Same as above.
• --output_dir: Same as above.
• --model_dir: Please specify the path of your directory where the pre-trained model weights are located.
• --rclass_list: Please specify list of RClass corresponding to sequential enzyme reactions of interest.
• --batch_size: Batch size in calculating probabilities. default:16
• --cuda: Same as above. default:False
• --cpu_num: Same as above. default:1

Evaluate results

python evaluate.py \
    --input_dir data \
    --output_dir output \
    --rclass_list RC01053 RC00004,RC00014 RC01923

• --input_dir: Same as above.
• --output_dir: Same as above.
• --rclass_list: Same as above.
• --window_size: Range of contiguous genes to be evaluated at a time. default:10
• --threshold: Threshold to obtain candidate genes. default:0.99
• --duplication: Whether to allow a single gene to be associated with multiple enzyme reactions. default:False

Test run

You can test DeepES by running the above three commands in sequence. After the testrun, the following directory structure is created.

output
├── candidate_genes
│   └── sample_data.tsv
├── embedding_vector
│   └── sample_data.pt
├── gene_table
│   └── sample_data.tsv
├── inference
│   ├── sample_data_RC00004.npy
│   ├── sample_data_RC00014.npy
│   ├── sample_data_RC01053.npy
│   └── sample_data_RC01923.npy
└── mapping_result
    └── sample_data.pkl

The result file is output/candidate_genes/sample_data.tsv:

score	window_idx	RC01053	RC00004,RC00014	RC01923
0.9999939003858481	0	eco:b2261	eco:b2260	eco:b2262
0.9999939003858481	1	eco:b2261	eco:b2260	eco:b2262
0.9999939003858481	2	eco:b2261	eco:b2260	eco:b2262
0.9999939003858481	3	eco:b2261	eco:b2260	eco:b2262
0.9999939003858481	4	eco:b2261	eco:b2260	eco:b2262
0.9999939003858481	5	eco:b2261	eco:b2260	eco:b2262

License

DeepES is released under the MIT License.