This document discusses running MUMmer within SyMAP and clustering the hits.
The corresponding MUMmer documentation is v3 and v4.

Disk space and timing

• The time and memory for MUMmer depends on the genomes sizes, complexity and similarity.

• If MUMmer fails, it is often due to insufficient memory; see Out of Memory, which explains how to determine the problem and ways around it.

• For large genomes, it is essential that the machine has at least 6Gb of RAM and disk space for each CPU used.

• If gene discovery is not important, then masking all but the gene sequences greatly reduces the time and memory. Alternatively, create a Blast file for input.

• When MUMmer is running, the temporary files can take up to 6G of disk space, hence, if you have 12 CPUs/threads running, this could use 72G of disk space.

MUMmer produces a .mum and .delta file. SyMAP only uses the .mum file so it removes the .delta file.
If you do not want SyMAP to remove the .delta, start symap with the "-mum" command line argument.

The following are sizes of MUMmer result data/seq_results/<species1>-to-<species2>/align directory:

You can remove the data/seq_results/<species1>-to-<species2> directory after SyMAP has finished the synteny computation, but it is strongly recommended that you leave them if you have the space. There are frequent SyMAP updates with improvements to the clustering and synteny computations; if you have kept these files, then you can update the synteny in very little time as SyMAP will reuse the MUMmer files (e.g. Hsa x Mus took >30 hours for MUMmer versus ~5 minutes for the synteny). Also, if you have saved the MUMmer files, you can easily try different parameters for the clustering and synteny stages.

The alignments can be compressed while not in use. The following was executed from the data/seq_results/Hsa_to_Pan directory, where the /align files took 12G of disk space:

tar -czf align.tar.gz align
rm -rf align				# this removes the directory

See Tested platforms and timing results

Use MUMmer

Preparing the sequences

Concat checked: • For the 1st genome, sequences are concatenated into a file as long as the file length is <1G. Multiple files of maximum file length 1G may be created. • For the 2nd genome, the same as above except the maximum file length is 60M. • All files from the 1st genome are searched against all files from the 2nd genome. This results in fewer MUMmer alignments, which can be faster. Concat unchecked: To reduce memory usage, you can uncheck Concat so that multiple files <60M are created for each genomes. This results in more MUMmer alignments, which can be slower. The exception is self-synteny, where all chromosomes are written to their own file, so Concat is not relevant.

The following statistics are from comparing Arabidopsis thaliana (119M) against Brassica rapa (297M) on a MacOS using 1 CPU.

48819 hits
  334 synteny blocks
46319 gene hits
38334 synteny hits

Finished in 1 hour 8 minutes

48846 hits
  334 synteny blocks
46348 gene hits
38345 synteny hits

Finished in 1 hour 35 minutes

Mask <abbrev>

MUMmer parameters

All MUMmer files but the those with the .mum suffix are removed by symap. If you prefer them not to be removed, use the "-mum" command line parameter, i.e.

 ./symap -mum

Existing MUMmer Files

Cluster MUMmer Hits

Algo1 vs Algo2

Algo1 is the default for self-synteny and if there is no annotation; else Algo2 is the default.

Wrong strand The wrong strand is when all hits in a cluster are to the same strand (++/--) yet the cluster aligns to two genes on the different strands (+-/-+), or vice versa.

Algo1 includes these hits. You can view them in the Queries where the Hit St column will be different than the two gene Gst columns.

Algo2 does NOT include these hits. You can request to view the potential hits during the A&S by running it with the "-wsp" flag, i.e. ./symap -wsp . This will only show gene pairs with (1) multiple hits to exons (in one or multiple gene pairs), (2) at least one is not an overlapping gene. It is up to the user to determine what is real.

Hints about parameter settings

Hint for Algo1: Increasing the Top N parameter can cause too many hits and reduce synteny. Decreasing it can remove more gene-pair hits. Hence, try Algo2 if you want more gene pairs.

Hint for Algo2: On the output to the terminal (in Verbose mode), if any chromosome pair shows over 10,000 hits, the parameters probably need to be made more stringent. Too many hits confuses the synteny algorithm, which results in synteny blocks not being found; it also results in very long execution time.

Suggestion: For large genomes, experiment with the parameters on just one pair of the chromosomes. (You can use xToSymap for the split.)

I have experimented with the datasets: (1) human, chimpanzee, mouse (2) Arabidopsis, Brassica rapa, Brassica oleracea. Only B. rapa to B. oleracea needed parameter adjustment: the number of G1 hits was over 200k, which is way more than typical; by increasing all parameters a small amount, this reduced to just over 100k.

Parameter description

Defaults: Algorithm 1 (Algo1): If at least one project does not have annotation or it is self-synteny. Algorithm 2 (Algo2): If both projects have annotation and it is not self-synteny.

3.c.iii Pseudos and Piles

The end of a hit may not overlap an annotated gene; by default, this will just show a Gene# of 'N.~' where N is the chromosome number. If Number Pseudo is selected, a pseudo Gene# will be assigned. The counts start after the annotated gene numbers and are suffixed by "~". For example, if the last Gene# for Chr03 is 5550 (e.g. 3.5550.), the first pseudo gene number will be 6000 (e.g. 3.6000.~).

If A&S was run without numbered pseudos, go to the Pair Parameter panel, and select Number Pseudo in the lower Drop-down; only this algorithm will be run. This cannot be undone; you would need to re-run A&S with the Number Pseudo unchecked to remove them.

Piles of Hits

The below image shows a pile of hits on the left (Cabb C5) that link to repetitive genes on the right (Arab Chr02). These are important to keep. The right image shows a pile of hits in an intergenic region (Cabbage Chr03) to multiple other regions (B.rapa Chr01). There are MANY occurrence of repeats like this in the MUMmer file, which is why these piles must be filtered; if they are not, the synteny algorithm does not perform well.

Email: cas1@arizona.edu