PLLM and SLLM Latent Protein Languages for Autoregressive Generation
October 1, 2026
Two new latent protein languages, PLL and SLL, enable more effective autoregressive generation of protein sequences and structures. PLL maps sequences to a 4,096-state contextual alphabet using ESM-2, while SLL uses a VQVAE-based approach to map to backbone coordinates.
HOW THIS AFFECTS YOU
●
researcherYou can use these learned latent spaces to improve transformer-based protein modeling performance.
●
healthThese advancements could accelerate de novo protein design and structural biology research.