bookmark ROME / Refinement and Optimization based on Machine lEarning for cryo-EM Offers a clustering method for single-particle cryo-EM data. ROME enables efficient computation of thousands of reference-free class averages in a highly affordable fashion. It can markedly improve the quality and resolution of ab initio 3D models with angular reconstitution. This tool was evaluated using several cryoEM datasets. It shows the advantage of generative topographic mapping (GTM)-based unsupervised clustering in discerning subtle structural differences directly from 2D class averages corresponding to distinct conformations.