Discrete Diffusion with Physical Mass Constraints for \emph{De Novo} Peptide Sequencing

$\textit{De novo}$ peptide sequencing is a pivotal technique that directly reconstructs amino acid sequences from tandem mass spectrometry (MS/MS) data; it enables the identification of novel proteins and variants absent from reference databases. Previous methods are typically based on autoregressive (AR) decoding or one-shot generation. The AR-based methods conflict with the bidirectional and globally constrained nature of MS/MS evidence and inevitably accumulate errors, while one-shot generation does not explicitly enforce physical constraints, failing to produce chemically valid and reliable peptides in a single pass. Accurate sequencing necessitates reasoning over the entire peptide simultaneously, enabling iterative self-correction under global constraints. To this end, we introduce $\textbf{PhysNovo}$, a novel paradigm that harnesses discrete diffusion to enable simultaneous global reasoning and iterative refinement. Specifically, PhysNovo reformulates sequencing as a $\textbf{phys}$ically mass-constrained reasoning process by embedding a knapsack-based feasibility kernel to enforce exact precursor mass consistency. By conditioning the diffusion process on global spectral context, PhysNovo supports abductive reasoning where bidirectional evidence is exploited to iteratively resolve local inconsistencies and ensure physically valid predictions. PhysNovo achieves state-of-the-art performance, exceeding baselines by over 2% in precision, with larger gains on out-of-distribution data. The source code is publicly available at https://github.com/WanyuGroup/ICML2026_PhysNovo.

De novo peptide sequencing is a technique that reconstructs protein fragments directly from mass spectrometry data, enabling scientists to discover new proteins, mutations, and disease-related biomarkers that are missing from existing databases. However, current AI systems often make predictions one amino acid at a time or generate sequences in a single pass, which can produce chemically impossible results and accumulate errors during decoding. In this work, we introduce PhysNovo, a new AI framework that treats peptide sequencing as an iterative reasoning process rather than a simple translation task. PhysNovo uses a discrete diffusion model that repeatedly refines its predictions while enforcing the physical laws of mass conservation measured by the instrument. The model also analyzes the entire spectrum globally, allowing it to detect and correct inconsistent amino acid assignments during generation instead of relying on post-processing fixes. Across multiple benchmark datasets, PhysNovo achieves state-of-the-art performance and shows particularly strong robustness on noisy spectra and previously unseen species. By integrating physical constraints directly into generative AI, our work demonstrates a more reliable and trustworthy approach for large-scale proteomics and biological discovery.