Dynamic programming algorithm to predict the optimal structure of multiple interacting RNA strands.

To install, goto mrnai-opt, and write make all

To run any program, you have to provide sequence information (sequences, their names, and their parity) to stdin. Since this will be needed for multiple programs, we recommend saving it in a file. All sequences should be provided in 5'->3'. For example, suppose the file yeast_trunc contains

NNNNGUAUGUNNNN&NNNNACAGAGAUGAUCAGCNNNN&NNNNGCUUAGAUCAAGUGUAGUANNNN&NNNNUACUAACACCNNNN
I1&U6&U2&I2
1,3|0,2

in the first three lines. The first line contains all sequences, separated by &. The second line has their names. The third line contains all even sequences' numbers followed by a |, followed by all odd sequences' numbers. Your file can contain other metadata as long as the first three lines are in this format.

Note that RNAs are layered in the order specified in line 1. To find the optimal structure from scratch, run

./run.sh inputs/yeast_trunc

assuming yeast_trunc is in the folder inputs. Results are written to stdout. The core program that is run here is findone. To see the full list of arguments that can be passed, execute build/findone -h.

Running this script generates several output files in the folder output. Most of them are intermediary files that contain output from RNAup, and have names such as default-X_Y_itemized.out, where X is an even RNA and Y is an odd RNA. A file partial_matrix.out will be created by the DP algorithm's program. This file contains the dynamic programming matrix, W(i_1, i_2, ..., i_m) for all indices. This file is needed by the suboptimal-enumeration algorithm's implementation. Finally, a file compiled_weights.out will be created, which contains data from all default-X_Y_itemized.out files, taking care of orientation information where necessary. This can be used with sampling programs.

The script ./run_convg can be used with the same arguments to heuristically converge to an optimal ordering of RNAs. The initial ordering doesn't matter, since the program randomizes the seed configuration every time it is run.

To take a brute force approach, generate structures for all permuations of RNAs. Use the script ./run_allperms.sh.

The program to generate all suboptimal solutions lying within the range [OPT, OPT(1+epsilon)] is also available in this package. As mentioned above, this program needs the partial dynamic programming matrix in order to run. To run everything from scratch (e.g. with epsilon = 0.1), use

./run_subopt.sh inputs/yeast_trunc 0.1

This will write all suboptimal structures and their energies to a file subopt_structs.out. If you already have the partial matrix, or prefer writing to another file, run the file subopt-enumeration/subopt_enum.py using the corresponding flags. Note that subopt_enum.py needs &-separated sequences from stdin.