Generative network architecture that may be used to produce de-novo protein sequences.

De novo protein design for catalysis of any desired chemical reaction is a long standing goal in protein engineering, due to the broad spectrum of technological, scientific and medical applications. Currently, mapping protein sequence to protein function is, however, neither computationionally nor experimentally tangible. Here we developed ProteinGAN, a specialised variant of the generative adversarial network that is able to 'learn' natural protein sequence diversity and enables the generation of functional protein sequences. ProteinGAN learns the evolutionary relationships of protein sequences directly from the complex multidimensional amino acid sequence space and creates new, highly diverse sequence variants with natural-like physical properties. Using malate dehydrogenase as a template enzyme, we show that 24% of the ProteinGAN-generated and experimentally tested sequences are soluble and display wild-type level catalytic activity in the tested conditions in vitro, even in highly mutated (>100 mutations) sequences. ProteinGAN therefore demonstrates the potential of artificial intelligence to rapidly generate highly diverse novel functional proteins within the allowed biological constraints of the sequence space.

All material is made available under Creative Commons BY-NC 4.0 license. You can use, redistribute, and adapt the material for non-commercial purposes, as long as you give appropriate credit by citing our paper and indicating any changes that you've made.

• Operating System: Linux.
• 64-bit Python 3.7 installation.
• blastp: 2.6.0+
• TensorFlow 1.13.1 or newer with GPU support.
• One or more NVIDIA GPUs. Recomendation: NVIDIA at least P100 GPU with 16GB.
• NVIDIA driver 418.87 or newer, CUDA toolkit 10.1 or newer, cuDNN 7.6.2 or newer.

environment.yml contains all the dependencies required in order to run ProteinGAN. You can simply run:

conda env create --file environment.yml

ProteinGAN expects a number of files in order to be able to train and evaluate the network.

Once data is ready, you can train your own ProteinGAN for chosen set of sequences as follows:

1. Edit gan/parameters.py to specify the dataset and training configuration.
2. Run the training script with python train_gan.
3. The results, weights will be stored in specified location. This location is printed once training script is executed. You can use tensorboard to view all the details.
4. The training may take several days (or weeks) to complete, depending on the configuration.
5. Once training is completed, you can use generate.py to generate chosen number of sequences.
6. Once training is completed, you can use discriminator_scores.py to get discriminator scores for all provided sequences.
7. Once training is completed, you can use test_gan.py to investigate GAN performance via interpolation.

• Database of annotated enzymes by its function - http://www.uniprot.org/.
• Database of enzyme reactions: https://www.expasy.org/.
• Paper on generating DNA sequences using GANs - https://arxiv.org/pdf/1712.06148.pdf
• Paper on generating peptides: https://arxiv.org/pdf/1804.01694.pdf

Papers influenced final solution:

• An Empirical Evaluation of Generic Convolutional and Recurrent Networksfor Sequence Modeling: https://arxiv.org/pdf/1803.01271.pdf
• Large Scale GAN Training for High Fidelity Natural Image Synthesis: https://arxiv.org/pdf/1809.11096.pdf
• Progressive Growing of GANs for Improved Quality, Stability, and Variation: https://arxiv.org/pdf/1710.10196.pdf
• Spectral Normalization for Generative Adversarial Networks: https://arxiv.org/abs/1802.05957.pdf
• Improved Techniques for Training GANs: https://arxiv.org/pdf/1606.03498.pdf
• Spectral Normalization for Generative Adversarial Networks: https://arxiv.org/pdf/1802.05957.pdf
• Multi-Scale Context Aggregation by Dilated Convolutions: https://arxiv.org/pdf/1511.07122.pdf
• Self-Attention Generative Adversarial Networks: https://arxiv.org/pdf/1805.08318.pdf
• cGANs with Projection Discriminator: https://arxiv.org/pdf/1802.05637.pdf
• A Style-Based Generator Architecture for Generative Adversarial Networks: https://arxiv.org/pdf/1812.04948.pdf
• Which Training Methods for GANs do actually Converge? https://arxiv.org/pdf/1801.04406