# How to Hallucinate Functional Proteins

**Authors:** Zak Costello, Hector Garcia Martin  
**arXiv:** [1903.00458](https://arxiv.org/abs/1903.00458)  
**Date:** 2019-03-01  
**Summary page:** https://www.summarizepaper.com/en/paper/1903.00458v1/

## Key points

- Authors present a novel approach to protein design and phenotypic inference using a generative model for protein sequences
- Introduce BioSeqVAE, a variational autoencoder variant that generates syntactically valid protein sequences likely to fold and function
- Model is trained on the entire known protein sequence space in an unsupervised manner
- Validation of BioSeqVAE's effectiveness through accurate reconstruction of sequences in its latent feature space
- Showcasing usefulness through various downstream design tasks
- Computational first step towards developing a general-purpose structure-free protein design tool
- Provides valuable insights into the potential applications of generative models in protein design

## Summary

In their paper titled "How to Hallucinate Functional Proteins," authors Zak Costello and Hector Garcia Martin present a novel approach to protein design and phenotypic inference using a generative model for protein sequences. They introduce BioSeqVAE, a variational autoencoder variant which has the ability to generate syntactically valid protein sequences that are likely to fold and function. The model is trained on the entire known protein sequence space in an unsupervised manner. The authors validate the effectiveness of BioSeqVAE by demonstrating that its latent feature space is useful and accurately reconstructs sequences. They also showcase its usefulness through various downstream design tasks. This work serves as a computational first step towards developing a general-purpose structure-free protein design tool. Overall, this research introduces an innovative method for generating functional protein sequences and provides valuable insights into the potential applications of generative models in protein design.