Coarse-Grained Model

The process of mapping groups of heavy atoms (e.g., a nucleotide base) into a single interaction site or bead. This reduces the number of degrees of freedom from thousands of atoms to hundreds of beads. The mapping is defined by selecting specific atoms as CG sites and calculating effective interactions that reproduce the thermodynamic properties of the all-atom reference system, a technique central to the MARTINI and SPICA force fields.

The resolution of a CG model dictates its accuracy and speed:

• One-Bead Models: Represent each nucleotide as a single interaction site, typically centered on the C3' or C4' atom. Extremely fast for long-timescale folding but cannot capture helical grooves or base-pair geometry.
• Three-Bead Models: Represent each nucleotide with distinct sites for the phosphate, sugar, and base. This captures directional hydrogen bonding and enables the formation of realistic secondary and tertiary structures, as implemented in the SimRNA and IsRNA tools.

CG models replace explicit electrostatic and van der Waals calculations with knowledge-based potentials or statistical potentials. These are derived by inverting the Boltzmann relation from a database of known RNA structures. For example, the DFIRE-RNA potential extracts pairwise distance-dependent pseudo-energies that implicitly account for solvation and counterion effects, guiding the CG chain toward native-like conformations without simulating water molecules.

CG models are frequently coupled with Replica Exchange Molecular Dynamics (REMD) to overcome rugged free-energy landscapes. Multiple copies of the system run in parallel at different temperatures, periodically attempting to swap coordinates. High-temperature replicas cross energy barriers, while low-temperature replicas refine local structure. The CG representation makes each replica simulation orders of magnitude faster, enabling convergence of the thermodynamic ensemble.

CG models provide an ideal framework for incorporating sparse experimental data as harmonic restraints or pseudo-energy terms. SHAPE reactivity data can be mapped onto CG beads to penalize conformations where a flexible nucleotide is buried. Similarly, cryo-EM density maps can be converted to a coarse-grained potential that pulls CG sites into high-density regions, driving the model toward the experimentally observed conformation during a folding simulation.

After a CG simulation identifies a stable folded state, the low-resolution structure must be reconstructed into an all-atom model for validation. Backmapping algorithms use fragment libraries or geometric rules to place atomic details onto the CG scaffold. Tools like CGTools or the SIRAH framework perform this reverse transformation, followed by short all-atom energy minimization to relieve local steric clashes and produce a physically realistic 3D structure.

The specific mapping of atomic groups to interaction sites defines the model's resolution. Common strategies include:

• One-bead models: Each nucleotide is a single sphere centered on the C3' or P atom, maximizing speed for long timescale folding.
• Three-bead models: Separate sites for phosphate, sugar, and base capture helical geometry and base-pairing directionality.
• Five-bead models: Additional sites for backbone torsional specificity, enabling discrimination between A-form and B-form helices. The choice trades chemical detail for computational tractability.

Coarse-grained models rely on statistical energy functions derived from known structures rather than physics-based force fields. These potentials capture:

• Base-pairing propensities: Frequencies of Watson-Crick, Hoogsteen, and Sugar-edge interactions from the Leontis-Westhof classification.
• Distance-dependent pair potentials: Radial distribution functions between bead types extracted from the PDB.
• Torsional statistics: Dihedral angle preferences for backbone and glycosidic bonds. The Boltzmann inversion technique converts observed probabilities into effective free energies.

The primary sampling method for coarse-grained RNA folding uses temperature cycling to escape local minima:

• High-temperature phase: Beads move freely, breaking incorrect base pairs and exploring global topologies.
• Cooling schedule: Gradual reduction of kinetic energy allows the system to settle into native-like contacts.
• Replica exchange: Multiple copies run in parallel at different temperatures, swapping configurations to enhance crossing of energy barriers. This approach is central to Rosetta FARFAR2 and SimRNA.

Coarse-grained trajectories must be converted back to atomic detail for validation and refinement. This process involves:

• Fragment assembly: Matching CG bead positions to the closest atomic fragments from a library of known structures.
• Energy minimization: Short all-atom relaxation to resolve steric clashes and optimize hydrogen bonding.
• Backbone closure: Ensuring continuous covalent connectivity across the reconstructed sugar-phosphate chain. The quality of reconstruction directly impacts final RMSD and TM-score metrics.

A widely used coarse-grained method employing a three-bead nucleotide representation and a knowledge-based statistical potential. Key features:

• Uses replica exchange Monte Carlo for efficient conformational sampling.
• Integrates SHAPE reactivity data as per-nucleotide restraints to guide folding.
• Outputs an ensemble of decoys ranked by energy, requiring clustering to identify the dominant conformation. SimRNA consistently performs well in RNA-Puzzles blind assessments.

Alternative coarse-grained frameworks that incorporate higher-resolution physics:

• IsRNA: A two-bead model (phosphate and nucleobase) with implicit solvent electrostatics, designed for simulating RNA-ligand and RNA-protein interactions.
• HiRE-RNA: A hierarchical approach using five beads per nucleotide with explicit hydrogen-bonding terms, enabling the study of protonation-dependent folding and catalytic mechanisms. Both bridge the gap between minimal CG models and explicit solvent molecular dynamics simulations.