Deep learning enables accurate design of functional transmembrane proteins

NeMO is a conditional deep learning model for environment-aware protein sequence design. By employing specific condition tags during inference, the model allows users to strictly dictate whether the generated sequence is tailored for a transmembrane or water-soluble context. Installation: conda create -n nemo python=3.10 conda activate nemo pip install torch torchvision fair-esm biotite biopython wandb pandas scipy Usage: Navigate to the examples/ directory and run the shell scripts for different design tasks: cd examples BR_TMP.sh / BR_WSP.sh: Monomer design with fixed residues (transmembrane / water-soluble) WSHC6_TMP.sh / WSHC6_WSP.sh: Homo-oligomer design with tied chain positions (transmembrane / water-soluble) See README.md for detailed usage instructions. Acknowledgements: We adapted some code from ProteinMPNN , OpenFold and CodonMPNN. We thank the authors for their impressive work. 1. J. Dauparas et al., Robust deep learning–based protein sequence design using ProteinMPNN. Science 378, 49-56 (2022). DOI: 10.1126/science.add2187 2. Ahdritz, G., Bouatta, N., Floristean, C. et al. OpenFold: retraining AlphaFold2 yields new insights into its learning mechanisms and capacity for generalization. Nat Methods 21, 1514–1524 (2024). https://doi.org/10.1038/s41592-024-02272-z 3. Stark, Hannes, et al. "CodonMPNN for organism specific and codon optimal inverse folding." arXiv preprint arXiv:2409.17265 (2024).