Nucleic acid structure prediction

From Bioinformatics.Org Wiki

Jump to: navigation, search

As with protein structure prediction methods, nucleic acid structure prediction attempts to determine the native, in vivo structure of a given nucleic acid sequence. For the determinants of nucleic acid structure, we do not see the importance of a buried hydrophobic core. There is instead an an increased emphasis on hydrogen bonding (base-pairing and the formation of double stranded structures) and inter-base (nearest neighbor) interactions (particularly "stacking energies").

For all macromolecules, it is the minimization of the total energy that ultimately determines structure. Computations are performed where the energy cost is the sum of all physical determinants for all interactions (one component, such as a bond or atom, interacting with another in the same macromolecule). And the computations can be highly compute-intensive.

Relative to deoxyribonucleic acids (DNA's), ribonucleic acids (RNA's) are short, and the single strand can fold back upon itself, forming the characteristic stem-loop secondary structure (e.g., loops, bulges and junctions). Again, an RNA will take the conformation where the total energy of the final structure is minimized. Computations largely take into account base-pairing, rewarding the more stable, double-stranded stem structures, where base-pairs are formed, and penalizing the less stable, single-stranded loop structures, where no base-pairs are formed.

Some computations also take into account stacking energies, which contribute to tertiary structure (e.g., pseudoknots and hairpin interactions) in RNA's. There are 10 unique nearest neighbor pairs (AA, CC, GG, TT, AC, AG, AT, CG, CT, GT), each with its own energy value, and these energies become more or less favorable at certain temperatures (see also denaturation).

The computations for RNA structure prediction require optimization routines (lowest energy = optimal), and dynamic programming, genetic algorithms, neural networks and Hidden Markov Models have been employed.

Software

See also

Personal tools
Namespaces
Variants
Actions
wiki navigation
Toolbox