DeepNano

DeepNano is a tool for predicting nanobody-antigen interactions. This repository contains code and model weights for DeepNano-seq, DeepNano-site and DeepNano. More details can be found in our paper .

---

System requirements

python==3.9, transformers==4.27.4 (or <=4.30.2), biopython==1.78, pandas==1.3.5, scikit-learn==1.0.2

---

Setup of DeepNano

The code was executed under python=3.9 and torch=1.13.1+cu116, we recommend you to use similar package versions.

Install DeepNano:

git clone https://github.com/ddd9898/DeepNano.git
cd DeepNano
pip install -r requirements.txt

This may take several minutes depending on your internet conditions.

---

Checkpoints

Our trained models can be downloaded at link .

DeepNano-seq(PPI)	DeepNano-seq(NAI)	DeepNano-site	DeepNano(NAI)
8M	8M	8M	8M
35M	35M	35M	35M
150M	150M	150M	150M
650M	650M	650M	650M

---

Demo

1. Download the checkpoint of DeepNano-seq(NAI) 8M and put it in the ./output/checkpoint.

2. Organize your input information as the following format:

   a. Pair data:

   Nanobody-ID	Antigen-ID	Label

   b. Sequence data:

   >nb0
   MAQVQLLESGGGLVQPGGSLRLSCAASGVTITDEDMTRVRQAPGKGLEWVSSILNTGGSTYYADSVKGRFTISRDNSKNTLYLQMNSLRAEDTAVYYCAAVHEKAADMNFWGQGTLVTVSS
   >at0
   DHNPFISVEWLKGPILEATAGDELVKLPVKLAAYPPPEFQWYKDGKALSGRHSPHALVLKEVTEASTGTYTLALWNSAAGLRRNISLELVVNVPPQIHEKEASSPSIYSRHSRQALTCTAYGVPLPLSIQWHWRPWTPCKMFAQRSLRRRQQQDLMPQCRDWRAVTTQDAVNPIESLDTWTEFVEGKNKTVSKLVIQNANVSAMYKCVVSNKVGQDERLIYFYVTTHHHHHH

​ Then change the three variables (fasta_path,pair_path and output_path) in ./predict.py.

3. Run the script to get predictions:

   python predict.py --model 1 --esm2 8M &
   

   where the --model 0, 1, and 2 correspond to the DeepNano-seq(PPI),DeepNano-seq(NAI), and DeepNano(NAI) models, respectively.

   Prediction results will be saved at the output_path .

---

Inference time

For the human serum albumin (609aa), to predict its interactions with 1 million nanobodies, the elapsed time for both DeepNano-seq and DeepNano models is shown in the table below:

	8M	35M	150M	650M
DeepNano-seq	1.15	2.27	6.92	20.3
DeepNano	2.30	4.62	13.92	40.8

Time units are in hours.

---

Reproduction

1. Get predictions on five PPI test datasets:

   python test_ppi_ESM2.py

2. Get predictions on the NAI test dataset:

   python test_nai.py

3. Virtual screening of anti-HSA and anti-GST:

   python test_case.py
   python test_background.py  --size 100w

---

Cite

Deng, J., Gu, M., Zhang, P. et al. Nanobody–antigen interaction prediction with ensemble deep learning and prompt-based protein language models. Nat Mach Intell (2024). https://doi.org/10.1038/s42256-024-00940-5

---

Contact

Feel free to contact [email protected] if you have issues for any questions.