analiza aiqbench
HD-Prot introduces a novel hybrid diffusion framework that integrates high-fidelity continuous structural tokens directly into protein language models without resorting to lossy discretization. By unifying discrete sequence prediction with continuous structure generation in a single architecture, this approach achieves state-of-the-art performance on tasks like inverse folding and motif-scaffolding while requiring significantly fewer computational resources than current multimodal baselines. This advancement represents a meaningful step forward in the efficiency and fidelity of generative models for biological data representation.
HD-Prot: A Protein Language Model for Joint Sequence-Structure Modeling with Continuous Structure Tokens
arXiv:2512.15133v3 Announce Type: replace-cross Abstract: Proteins inherently possess a consistent sequence-structure duality. The abundance of protein sequence data, which can be readily represented as discrete tokens, has driven fruitful developments in protein language models (pLMs). A key remaining challenge, however, is how to effectively integrate continuous structural knowledge into pLMs. Current methods often discretize protein structures to accommodate the language modeling framework, which inevitably results in the loss of fine-grained information and limits the performance potential of multimodal pLMs. In this paper, we argue that such concerns can be circumvented: a sequence-based pLM can be extended to incorporate the structure modality through continuous tokens, i.e., high-fidelity protein structure latents that avoid vector quantization. Specifically, we propose a hybrid diffusion protein language model, HD-Prot, which embeds a continuous-valued diffusion head atop a discrete pLM, enabling seamless operation with both discrete and continuous tokens for joint sequence-structure modeling. It captures inter-token dependencies across modalities through a unified absorbing diffusion process, and estimates per-token distributions via categorical prediction for sequences and continuous diffusion for structures. Extensive results demonstrate that HD-Prot achieves competitive performance in unconditional sequence-structure co-generation, motif-scaffolding, protein structure prediction, and inverse folding tasks. Furthermore, our method can perform on par with state-of-the-art multimodal pLMs, despite being developed under limited computational resources (i.e., less than one-tenth the budget for modality extension fine-tuning). It highlights the viability of simultaneously estimating categorical and continuous distributions within a unified language model architecture, offering a promising alternative direction for multimodal pLMs.
Abstract:Proteins inherently possess a consistent sequence-structure duality. The abundance of protein sequence data, which can be readily represented as discrete tokens, has driven fruitful developments in protein language models (pLMs). A key remaining challenge, however, is how to effectively integrate continuous structural knowledge into pLMs. Current methods often discretize protein structures to accommodate the language modeling framework, which inevitably results in the loss of fine-grained information and limits the performance potential of multimodal pLMs. In this paper, we argue that such concerns can be circumvented: a sequence-based pLM can be extended to incorporate the structure modality through continuous tokens, i.e., high-fidelity protein structure latents that avoid vector quantization. Specifically, we propose a hybrid diffusion protein language model, HD-Prot, which embeds a continuous-valued diffusion head atop a discrete pLM, enabling seamless operation with both discrete and continuous tokens for joint sequence-structure modeling. It captures inter-token dependencies across modalities through a unified absorbing diffusion process, and estimates per-token distributions via categorical prediction for sequences and continuous diffusion for structures. Extensive results demonstrate that HD-Prot achieves competitive performance in unconditional sequence-structure co-generation, motif-scaffolding, protein structure prediction, and inverse folding tasks. Furthermore, our method can perform on par with state-of-the-art multimodal pLMs, despite being developed under limited computational resources (i.e., less than one-tenth the budget for modality extension fine-tuning). It highlights the viability of simultaneously estimating categorical and continuous distributions within a unified language model architecture, offering a promising alternative direction for multimodal pLMs.
Computational Engineering, Finance, and Science (cs.CE); Artificial Intelligence (cs.AI)
arXiv:2512.15133 [cs.CE]
(or arXiv:2512.15133v3 [cs.CE] for this version)
Text integral sub licența deschisă a publicației · 257 cuvinte
Subiecte conexe
Multimodal Co-Generation With Continuous Bitstream Diffusion
arXiv · 14 oct. 2026
LC-SEPLM: long-range contact-supervised adaptation for sequence-only protein representation learning
arXiv · 28 iul. 2026
ProTok: Unified Protein Tokenization via Modular Multi-Codebook Quantization
arXiv · 14 oct. 2026
News
Quantum Computing Report · 25 iul. 2026