Simulation Based Inference for Multi-component Nanoparticle Scattering

Small-angle X-ray scattering (SAXS) enables rapid structural characterization of nanoparticles, but inferring physical parameters from multi-component scattering profiles is complicated by overlapping contributions and degeneracies. We develop a simulation-based inference (SBI) framework that uses a masked autoregressive flow trained on synthetic SAXS profiles from a multi-component forward model. On held-out simulations, the model achieves accurate parameter recovery and well-calibrated posterior uncertainties. Applied to experimental SAXS measurements, SBI produces reconstructions with accuracy comparable to independent blind least-squares fits while additionally providing posterior uncertainty and information about possible degeneracies. These results demonstrate the potential of amortized SBI for fast, uncertainty-aware SAXS analysis.