MAGNA++ runs on Linux, Mac OS X, and Windows, and is available for download as a compressed file here. Statically compiled executables are provided for all three operating systems, so the aligner is ready to run immediately after downloading, though instructions are also provided to compile directly from the source.

The suggested network size for MAGNA++ is up to 10,000 nodes and 50,000 edges. The aligner's memory requirement increases linearly with respect to the number of nodes and edges. The time taken to run the aligner increases linearly with respect to the number of edges, with the running time decreasing close to linearly with the number of cores used.

The MAGNA++ download takes up 1-2MB of disk space, though more space will be required to store networks and the resulting alignments between them.

Installing involves simply uncompressing the downloaded file and running MAGNA++.

Start MAGNA++ by running the executable. MAGNA++ comes with a user-friendly graphical interface; below is a screenshot of the interface along with descriptions of each input field.

I.
Choose two networks to align. Network 1 must have fewer nodes than Network 2. MAGNA++ accepts networks in either LEDA network format (.gw) (see an example), edge list format (.txt) (see an example) or Simple Interaction Format (.sif) (see an example). Various real-world, synthetic, testing, and toy networks (LEDA formatted) can be found here.

II.
Choose the output directory and a prefix for each output file name. For example, if "ex_" is entered, then "ex_" will appear at the beginning of the name of each file that MAGNA++ outputs. MAGNA++ will output files for the resulting alignments, the parameters used to create the alignments, and some statistics of the alignments. See Section 3 for a more detailed description of output files.

III.
Choose an edge-based measure to optimize. S³ is the recommended optimization measure since it avoids biases due to network sizes.

IV.
Choose whether to optimize a node-based measure in addition to optimizing an edge-based measure. If "None" is selected, then MAGNA++ will only optimize edge conservation (see IV). If "Node Similarity File" is selected, then choose a node similarity file. MAGNA++ can read node similarity files in two formats: the simple matrix format (see an example) or the CSV matrix format (see an example).

Supposing Network 1 has m nodes and Network 2 has n nodes. The simple matrix format should contain the size of the matrix on the first line and then m lines and n columns delimited by spaces. The CSV matrix format should have m x n lines, the first column containing the nodes in Network 1, the second the nodes in Network 2, and the third column the similarity value between the two nodes.

V.
Choose a parameter . Using this parameter, MAGNA++ can simultaneously optimize an edge-based measure (EC, ICS, or S³) and a node-based measure (of topological or biological node similarities). In particular, given an edge-based measure , a node-based measure , and a parameter between 0 and 1, MAGNA++ will optimize . When is 1, MAGNA++ will only optimize edge conservation, and when , MAGNA++ will only optimize node conservation.

VI.
MAGNA++ can either improve an existing set of alignments or start from scratch. Choose either "None" or "File". If "File" is selected, then MAGNA++ will prompt for an initial population file, which will define an existing set of alignments for which to improve. This file should contain a list of alignments file names (see an example), which must be in the same directory as the initial population file. If "None" is selected, no initial population will be used, so MAGNA++ will start from scratch by generating a random set of initial alignments.

VII.
MAGNA++ uses a genetic algorithm, for which there are three parameters:

"Population Size" is the population size of the genetic algorithm; that is, the genetic algorithm will work with this many alignments with each iteration. If no initial population is provided (see VI), then 15,000 is generally the recommended population size. If using alignments generated by other algorithms (such as IsoRank) in the initial population, then a population size of 10,000 is recommended.

"Number of Generations" is the number of iterations or generations for which the genetic algorithm will run. If no initial population is used, then 2,000 is generally the recommended number of generations. If using an alignments generated by another algorithm, then 1,000 is the recommended number of generations.

"Fraction of Elite Members" is the highest-quality fraction of the population that survives to the next generation. 0.5 is always the recommended value.

VIII.
MAGNA++ is capable of using multiple threads to perform its computation. It parallelizes the calculation of alignment quality, which is the computational bottleneck of MAGNA++. Calculation of alignment quality is divided among threads. MAGNA++ automatically determines the number of threads by using as many threads as there are cores in the computer, though this value can be changed manually. However, using more threads than there are cores is not recommended since performance is not likely to improve. Performance will drop if too many threads are used.

IX.
Choose how often MAGNA++ outputs the best alignment in the population. If the frequency of output is k, and there are n generations, then MAGNA++ will output the best alignment every n/k generations. For example, if k = 1, then MAGNA++ will output the best alignment with every generation, and if k = n, then MAGNA++ will only output the best alignment of the last generation. It is recommended that k divides n.

Instructions for running the command line interface (CLI) of MAGNA++ can be found in the README.txt file that accompanies the CLI executables of MAGNA++.

Upon beginning an execution (and supposing the prefix defined in 2.II is "ex_"), MAGNA++ generates three output files by default. The final alignment file "ex_final_alignment.txt" consists of the aligned pairs of nodes calculated by MAGNA++. The alignment statistics file "ex_final_stats.txt" contains the EC, ICS, S³, node similarity, and the overall score of the final alignment. The common conserved subgraph file, "ex_final_visualize.sif", contains the common conserved subgraph of the alignment of the two networks. It provides an intuitive visualization of the final alignment, and can be viewed using a program such as Cytoscape.

If the frequency of output option is set to k, then MAGNA++ outputs the details of the best alignment every n/k generations (where n is the total number of generations, asdefined in 2.XI). Each alignment is written to a file, containing a list of node pairs. Nodes in the left column are from Network 1 and the nodes on the right are from Network 2. Also according to the frequency of output, MAGNA++ will append statistics of the best alignment to the file "ex_stats.txt".

The format of the output alignment file names is "ex_M_p_n_i.aln", where M is the edge conservation measure (one of EC, ICS, S3), p is the population size, n is the total number of generations, and i is the generation that the alignment was written. The alignment from the last generation (ex_M_p_n_n.aln) is the final alignment produced by MAGNA++.

MAGNA++ will also create a file "ex_params.txt", which contains a summary of the initial parameters given.

MAGNA++ is created by V. Vijayan, V. Saraph and T. Milenkovic at University of Notre Dame.

MAGNA++ is freely available for academic use only.

Disclaimer: THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.