GenProSeq

GenProSeq generates novel protein sequences using deep generative models for protein engineering applications in synthetic biology, medicine, and material science.

Key Features:

• Variational Autoencoder (VAE): Generates protein sequences using Word2vec embedding to learn biologically meaningful representations.
• Generative Adversarial Networks (GAN): Generates protein sequences using Word2vec embedding to model the underlying distribution of protein data.
• Autoregressive Models: Generates protein sequences using the transformer encoder applied to protein sequences.
• Novel Sequence Generation: Produces protein sequences that have not been observed before and assigns higher probability to sequences that satisfy desired criteria.

Scientific Applications:

• Protein Engineering: Generates useful protein sequences on a variety of scales for engineering applications.
• Representation Learning: Learns biologically meaningful representations helpful for downstream tasks in protein analysis.
• De Novo Protein Design: Creates novel protein sequences with similar properties to training data for synthetic biology and medicine.
• Property-Guided Generation: Assigns higher probability to protein sequences that satisfy desired functional or structural criteria.

Methodology:

Deep generative models including variational autoencoder, generative adversarial networks, and autoregressive models are implemented; Word2vec is used for embedding in VAE and GAN, and the transformer encoder is applied to protein sequences for the autoregressive model.