# VQPL: Vector Quantized Protein Language

arXiv: https://arxiv.org/abs/2310.04985 · HTML: https://mlsanity.com/paper/2310.04985/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 8 Oct 2023.

## Built on by (most cited first, 4 of 4)

- [Masked Modeling for Self-supervised Representation Learning on Vision and Beyond](https://mlsanity.com/paper/2401.00897/) (2023)
- [Learning the Language of Protein Structure](https://mlsanity.com/paper/2405.15840/) (2024)
- [Efficiently Predicting Protein Stability Changes Upon Single-point Mutation with Large Language Models](https://mlsanity.com/paper/2312.04019/) (2023)
- [FoldToken2: Learning compact, invariant and generative protein structure language](https://mlsanity.com/paper/2407.00050/) (2024)

## Abstract (arXiv)

Is there a foreign language describing protein sequences and structures simultaneously? Protein structures, represented by continuous 3D points, have long posed a challenge due to the contrasting modeling paradigms of discrete sequences. To represent protein sequence-structure as discrete symbols, we propose a VQProteinformer to project residue types and structures into a discrete space, supervised by a reconstruction loss to ensure information preservation. The sequential latent codes of residues introduce a new quantized protein language, transforming the protein sequence-structure into a unified modality. We demonstrate the potential of the created protein language on predictive and generative tasks, which may not only advance protein research but also establish a connection between the protein-related and NLP-related fields. The proposed method will be continually improved to unify more protein modalities, including text and point cloud.