Cryo-EM Density Map

The dominant cryo-EM technique where 2D projection images of randomly oriented, identical macromolecules are computationally aligned and averaged to reconstruct a 3D density map. This method overcomes the low signal-to-noise ratio inherent in low-dose imaging by treating each particle image as a noisy observation of the same underlying signal. Key steps include:

• Motion correction: Compensating for beam-induced sample movement
• CTF estimation: Modeling the contrast transfer function to correct for microscope aberrations
• 2D classification: Grouping similar particle views to discard junk and heterogeneity

The computational process of placing an atomic model into a cryo-EM density map to maximize the correlation between the model's calculated scattering and the experimental reconstruction. For RNA, this involves:

• Rigid-body docking: Placing entire helical domains as fixed units into segmented density
• Flexible fitting: Using molecular dynamics flexible fitting (MDFF) to apply density-derived forces that pull atoms into the map while maintaining stereochemical plausibility
• Real-space refinement: Iteratively adjusting atomic coordinates and B-factors against the map using programs like Phenix

A method for quantifying the spatially varying quality within a single cryo-EM density map. Unlike global Fourier shell correlation (FSC) which reports a single resolution value, local resolution maps reveal:

• Flexible regions: Mobile domains like RNA loops often show lower resolution than rigid helical cores
• Preferred orientation artifacts: Anisotropic sampling creates directional resolution anisotropy
• Masking effects: Solvent-exposed regions may appear weaker due to disorder Tools like ResMap and MonoRes compute per-voxel resolution estimates used to weight map-to-model fitting restraints.

Post-processing operations that apply a negative temperature factor to restore high-resolution detail attenuated by the imaging process. The B-factor describes the exponential decay of structure factor amplitudes with increasing spatial frequency due to:

• Envelope function: Microscope optics and detector modulation
• Radiation damage: Progressive loss of high-resolution features during exposure
• Conformational heterogeneity: Residual structural variability in the particle ensemble Over-sharpening introduces noise amplification, so tools like phenix.auto_sharpen optimize the B-factor to balance detail and map interpretability.

Neural network approaches that improve cryo-EM map interpretability by learning to separate signal from noise without requiring paired clean targets. Architectures include:

• Noise2Noise: Training on pairs of independent half-maps to learn a denoising function without clean ground truth
• Topaz-Denoise: A convolutional neural network pre-trained on cryo-EM data for particle micrograph denoising prior to picking
• DeepEMhancer: A U-Net architecture that performs post-processing sharpening and local filtering to enhance map features for model building These methods preserve biologically relevant density while suppressing solvent noise and reconstruction artifacts.

Specialized quality assessment tools that go beyond global map-model correlation to validate RNA atomic models against cryo-EM density:

• Q-score: A per-atom metric quantifying how well each atom is resolved by the map, calculated from the similarity of the atom's density profile to a reference Gaussian
• Sukar score: Evaluates sugar pucker geometry and backbone dihedral conformity to known RNA rotamer distributions
• Clashscore: Identifies steric overlaps that indicate fitting errors, particularly important in tightly packed RNA tertiary folds
• Pperp distance: Measures deviation of base atoms from the plane defined by the purine or pyrimidine ring, sensitive to local fitting errors