PI-Mamba: linear-time protein backbone generation via spectrally initialized flow matching

Generative models for protein backbone design have to simultaneously ensure geometric validity, sampling efficiency, and scalability to long protein chains. However, most existing approaches rely on iterative refinement, quadratic attention mechanisms, or post-hoc geometry correction, leading to a persistent trade-off between computational efficiency and structural fidelity.

We present Physics-Informed Mamba (PI-Mamba), a generative model that enforces exact local covalent geometry by construction while enabling linear-time inference. PI-Mamba integrates a differentiable constraint-enforcement operator into a flow-matching framework and couples it with a Mamba-based state-space architecture. To improve optimization stability and backbone realism, we introduce a spectral initialization derived from the Rouse polymer model and an auxiliary cis-proline awareness head. Across benchmark tasks, PI-Mamba demonstrates the advantage in scalable, physically valid backbone generation: on a single A5000 GPU (24GB), it generates backbones beyond 2000 residues, producing 2000-residue samples in 9.49 s with only 0.91 GB peak VRAM, while preserving exact local geometry with 0.0% local geometry violations and maintaining strong designability on short-chain benchmarks (mean scTM = 0.910 at L = 100).

Code and distilled data are available at https://github.com/forxhunter/PI-mamba.