8 Mar 2012 mcl 12-068
mcl — The Markov Cluster Algorithm, aka the MCL algorithm.
This program implements mcl, a cluster algorithm for graphs. A single parameter controls the granularity of the output clustering, namely the -I inflation option described further below. In standard usage of the program this parameter is the only one that may require changing. By default it is set to 2.0 and this is a good way to start. If you want to explore cluster structure in graphs with MCL, vary this parameter to obtain clusterings at different levels of granularity. A good set of starting values is 1.4, 2, 4, and 6.
The program has a rather large set of options. Except for -I none affects the clustering method itself. The other options are for a variety of aspects, such as study of the underlying MCL process (i.e. dumping of iterands), network preprocessing (incorporated for efficiency), resource allocation options (for large-scale analyses), output naming and placement, output formatting, setting of verbosity levels, and so on.
Network construction and reduction techniques should not be considered as part of a clustering algorithm. Nevertheless particular techniques may benefit particular methods or applications. In mcl many transformations are accessible through the -tf option. It can be used for edge weight transformations and selection, as well as transformations that act on a graph as a whole. It is for example possible to remove edges with weight below 0.7 by issuing -tf 'gq(0.7)', where the quotes are necessary to prevent the shell from interpreting the parentheses. The option accepts more complicated sequences, such as -tf 'gq(0.7),add(-0.7)'. This causes all remaining edge weights to be shifted to the range [0-0.3], assuming that the input contains correlations. Many more transformations are supported, as documented in mcxio. Examples of graph-wide transformations are '#knn(<num>)' and '#ceilnb(<num>)'. The first only keeps those edges that occur in the list of top-<num> edges of highest weight in both of its incident nodes. The second removes edges from nodes of highest degree first, proceeding until all node degrees satisfy the given threshold. The -pi (pre-inflation) option can be used to increase the contrast in edge weights. This may be useful when clusterings are coarse and fine-grained clusterings are difficult to obtain.
There are two main modes of invocation. The most accessible is label mode which assumes a format alternatively called label input or ABC-format. The input is then a file or stream in which each line encodes an edge in terms of two labels (the 'A' and the 'B') and a numerical value (the 'C'), all separated by white space. The most basic example of usage is this:
The output is then a file where each line is a cluster of tab-separated labels. If clustering is part of a larger workflow where it is desirable to analyse and compare multiple clusterings, then it is a good idea to use native mode rather than ABC mode. The reason for this is that native mode is understood by all programs in the mcl suite. It is a more stringent and unambiguous format, and hence more suitable for data exchange. The reader is refered to clmprotocols for more information.
The example invocation below assumes matrix input, as explained above and described in the mcxio section. Switching to label mode requires the input file to be in ABC-format and the addition of the --abc option.
mcl <-|fname> [-I <num> (inflation)] [-o <str> (fname)]
These options are sufficient in 95 percent of the cases or more. The first argument must be the name of a file containing a graph/matrix in the mcl input format, or a hyphen to read from STDIN. With respect to clustering, the -I option is foremost relevant.
The full listing of mcl options is shown further below, separated into parts corresponding with functional aspects such as clustering, threading, verbosity, network preprocessing, pruning and resource management, automatic output naming, and dumping.
Baseline clustering options
[-I <num> (inflation)] [-o <fname> (fname)]
[-odir <dname> (directory)] [--d (use input directory for output)]
[--abc (expect/write labels)] [--sif (expect/write labels)] [--etc (expect/write labels)] [--expect-values (sif or etc stream contains values)] [-use-tab <fname> (use mapping to write)]
[-tf <tf-spec> (transform input matrix values)] [-abc-tf <tf-spec> (transform input stream values)] [--abc-neg-log10 (take log10 of stream values, negate sign)] [--abc-neg-log (take log of stream values, negate sign)]
[-write-graph <fname> (write graph)] [-write-graphx <fname> (write transformed graph)] [-write-expanded <fname> (write expanded graph)] [--write-limit (write mcl process limit)]
Input manipulation options
[-pi <num> (pre-inflation)] [-ph <num> (pre-inflation, max-bound)] [-if <num> (start-inflation)] [--discard-loops=<y/n> (discard y/n loops in input)] [--sum-loops (set loops to sum of other arcs weights)] [-c <num> (reweight loops)]
Clustering processing options
[-sort <str> (sort mode)] [-overlap <str> (overlap mode)] [--force-connected=<y/n> (analyze components)] [--check-connected=<y/n> (analyze components)] [--analyze=<y/n> (performance criteria)] [--show-log=<y/n> (show log)]
[-q <spec> (log levels)] [-v <str> (verbosity type on)] [-V <str> (verbosity type off)] [--show (print (small) matrices to screen)]
[-te <int> (#expansion threads)]
Output file name and annotation options
[-o <str> (fname)] [-ap <str> (use str as file name prefix)] [-aa <str> (append str to suffix)] [-az (show output file name and exit)] [-ax (show output suffix and exit)] [-annot <str> (dummy annotation option)]
[-dump-interval <i:j> (dump interval)] [-dump-modulo <int> (dump modulo)] [-dump-stem <stem> (dump file stem)] [-dump <str> (type)] [-digits <int> (printing precision)] [--write-binary (write matrices in binary format)]
[--jury-charter (explains jury)] [--version (show version)] [-how-much-ram k (RAM upper bound)] [-h (most important options)] [--help (one-line description for all options)] [-z (show current settings)] [-az (show output file name and exit)] [-ax (show output suffix and exit)] [--show-schemes (show resource schemes)]
[-sparse <int> (sparse matrix multiplication threshold)]
The following options all pertain to the various pruning strategies that can be employed by mcl. They are described in the PRUNING OPTIONS section, accompanied by a description of the mcl pruning strategy. If your graphs are huge and you have an appetite for tuning, have a look at the following:
[-scheme <int> (resource scheme)] [-resource <int> (per-node resource maximum)] [-p <num> (cutoff)] [-P <int> (1/cutoff)] [-S <int> (selection number)] [-R <int> (recovery number)] [-pct <int> (recover percentage)] [-warn-pct <int> (prune warn percentage)] [-warn-factor <int> (prune warn factor)]
The first argument of mcl must be a file name, but some options are allowed to appear as the first argument instead. These are the options that cause mcl to print out information of some kind, after which it will gracefully exit. The full list of these options is
-z, -h, --help, --version, --show-settings, --show-schemes, --jury-charter.
mcl implements the MCL algorithm, short for the Markov cluster algorithm, a cluster algorithm for graphs developed by Stijn van Dongen at the Centre for Mathematics and Computer Science in Amsterdam, the Netherlands. The algorithm simulates flow using two simple algebraic operations on matrices. The inception of this flow process and the theory behind it are described elsewhere (see REFERENCES). Frequently asked questions are answered in the mclfaq section. The program described here is a fast threaded implementation written by the algorithm's creator with contributions by several others. Anton Enright co-implemented threading; see the HISTORY/CREDITS section for a complete account. See the APPLICABILITY section for a description of the type of graph mcl likes best, and for a qualitative assessment of its speed. mcl is accompanied by several other utilities for analyzing clusterings and performing matrix and graph operations; see the SEE ALSO section.
The first argument is the input file name, or a single hyphen to read from stdin. The rationale for making the name of the input file a fixed parameter is that you typically do several runs with different parameters. In command line mode it is pleasant if you do not have to skip over an immutable parameter all the time.
The -I f option is the main control, affecting cluster granularity. In finding good mcl parameter settings for a particular domain, or in finding cluster structure at different levels of granularity, one typically runs mcl multiple times for varying values of f (refer to the -I inflation option for further information).
NOTE MCL interprets the matrix entries or graph edge weights as similarities, and it likes undirected input graphs best. It can handle directed graphs, but any node pair (i,j) for which w(i,j) is much smaller than w(j,i) or vice versa will presumably have a slightly negative effect on the clusterings output by mcl. Many such node pairs will have a distinctly negative effect, so try to make your input graphs undirected. How your edge weights are computed may affect mcl's performance. In protein clustering, one way to go is to choose the negated logarithm of the BLAST probabilities (see REFERENCES).
mcl's default parameters should make it quite fast under almost all circumstances. Taking default parameters, mcl has been used to generate good protein clusters on 133k proteins, taking 10 minutes running time on a Compaq ES40 system with four alpha EV6.7 processors. It has been applied (with good results) to graphs with two million nodes, and if you have the memory (and preferably CPUs as well) nothing should stop you from going further.
For large graphs, there are several groups of parameters available for tuning the mcl computing process, should it be necessary. The easiest thing to do is just vary the -scheme option. This triggers different settings for the group of pruning parameters -p/-P, -R, -S, and -pct. The default setting corresponds with -scheme 6. When doing multiple mcl runs for the same graphs with different -I settings (for obtaining clusterings at different levels of granularity), it can be useful to factor out the first bit of computation that is common to all runs, by using the -write-expanded option one time and then using -if inflation for each run in the set. Whether mcl considers a graph large depends mainly on the graph connectivity; a highly connected graph on 50,000 nodes is large to mcl (so that you might want to tune resources) whereas a sparsely connected graph on 500,000 nodes may be business as usual.
mcl is a memory munger. Its precise appetite depends on the resource settings. You can get a rough (and usually much too pessimistic) upper bound for the amount of RAM that is needed by using the -how-much-ram option. The corresponding entry in this manual page contains the simple formula via which the upper bound is computed.
Other options of interest are the option to specify threads -te, and the verbosity-related options -v and -V. The actual settings are shown with -z, and for graphs with at most 12 nodes or so you can view the MCL matrix iterands on screen by supplying --show (this may give some more feeling).
MCL iterands allow a generic interpretation as clusterings as well. The clusterings associated with early iterands may contain a fair amount of overlap. Refer to the -dump option, the mclfaq manual, and the clm imac utility (Interpret Matrices As Clusterings). Use clm imac only if you have a special reason; the normal usage of mcl is to do multiple runs for varying -I parameters and use the clusterings output by mcl itself.
Under very rare circumstances, mcl might get stuck in a seemingly infinite loop. If the number of iterations exceeds a hundred and the chaos indicator remains nearly constant (presumably around value 0.37), you can force mcl to stop by sending it the ALRM signal (usually done by kill -s ALRM pid). It will finish the current iteration, and interpret the last iterand as a clustering. Alternatively, you can wait and mcl might converge by itself or it will certainly stop after 10,000 iterations. The most probable explanation for such an infinite loop is that the input graph contains the flip-flop graph of node size three as a subgraph.
The creator of this page feels that manual pages are a valuable resource, that online html documentation is also a good thing to have, and that info pages are way way ahead of their time. The NOTES section explains how this page was created.
In the OPTIONS section options are listed in order of importance, with related options grouped together.
Sets the main inflation value to <num>. This value is the main handle for affecting cluster granularity. It is usually chosen somewhere in the range [1.2-5.0]. -I 5.0 will tend to result in fine-grained clusterings, and -I 1.2 will tend to result in very coarse grained clusterings. Your mileage will vary depending on the characteristics of your data. That is why it is a good idea to test the quality and coherency of your clusterings using clm dist and clm info. This will most likely reveal that certain values of -I are simply not right for your data. The clm dist section contains a discussion of how to use the cluster validation tools shipped with mcl (see the SEE ALSO section).
With low values for -I, like -I 1.2, you should be prepared to use more resources in order to maintain quality of clusterings, i.e. increase the argument to the -scheme option.
The default mode of output creation for mcl is to create a file name that uses the input file name stripped of any leading path components, augmented with a prefix 'out.' and a suffix encoding pivotal mcl parameters. This will usually be the inflation value which is the argument to the -I option. By default the output file is written in the current directory. For example, if the input is named data/small.mci for example and inflation is set to three, the output file will be named out.small.mci.I30.
This behaviour can be overridden in various ways. The -o option simply specifies the output file name, which may include path components that should exist. It is possible to send the clustering to STDOUT by supplying -o -. With the -odir <dname> option mcl constructs the output file name as before, but writes the file in the directory <dname>. Finally, the option --d is similar but more specific in that mcl will write the output in the directory specified by the path component of the input file, that is, the directory in which the input file resides.
If either one of --abc, --sif, --etc or -use-tab tab-file is used the output will be in label format. Otherwise the clustering is output in the mcl matrix format; see the mcxio section for more information on this. Refer also to the group of options discussed at --abc.
Look at the -ap prefix option and its siblings for the automatic naming constructions employed by mcl if the -o option is not used.
With the -c <num> option, as the final step of loop computation (i.e. after initialization and shadowing) all loop weights are multiplied by <num>, if supplied.
By default mcl will remove any loops that are present in the input. Use --discard-loops=n to turn this off. Bear in mind that loops will still be modified in all cases where the loop weight is not maximal among the list of edge weights for a given node.
These items all relate to label input and/or label output. --abc tells mcl to expect label input and output clusters in terms of those labels. This simple format expects two or three fields separated by white space on each line. The first and second fields are interpreted as labels specifying source and destination node respectively. The third field, if present, specifies the weight of the arc connecting the two nodes.
The option --sif tells mcl to expect SIF (Simple Interaction File) format. This format is line based. The first two fields specify the source node (as a label) and the relationship type. An arbitrary number of fields may follow, each containing a label identifying a destination node. The second field is simply ignored by mcl. As an extension to the SIF format weights may optionally follow the labels, separated from them with a colon character. It is in this case necessary to use the --expect-values option. The --etc option expects a format identical in all respects except that the relationship type is not present, so that all fields after the first are interpreted as destination labels.
-use-tab is only useful when matrix input is used. It will use the tab file to convert the output to labels; it does not fail on indices missing from the tab file, but will bind these to generated dummy labels.
-tf transforms the values of the input matrix according to <tf-spec>. -abc-tf transforms the stream values (when --abc is used) according to <tf-spec>. --abc-neg-log and --abc-neg-log10 imply that the stream input values are replaced by the negation of their log or log10 values, respectively. The reason for their existence is documented in mcxio. For a description of the transform language excpected/accepted in <tf-spec> refer to the same.
The first two options are somewhat outdated, in that the prefered way of loading networks is by using mcxload. The option -write-expanded can be useful for exploring more complicated input transformations that incorporate an expansion step, but is not really relevant for production use. The last option is mainly educational and for analyzing the mcl process itself.
There are currently seven different resource schemes, indexed 1..7. High schemes result in more expensive computations that may possibly be more accurate. The default scheme is 4. When mcl is done, it will give a grade (the so called jury synopsis) to the appropriateness of the scheme used. A low grade does not necessarily imply that the resulting clustering is bad - but anyway, a low grade should be reason to try for a higher scheme.
Use the -resource <num> option to cap for each nodes the number of neighbours tracked during computation at <num> nodes.
The PRUNING OPTIONS section contains an elaborate description of the way mcl manages resources, should you be interested. In case you are worried about the validation of the resulting clusterings, the mclfaq section has several entries discussing this issue. The bottom line is that you have to compare the clusterings resulting from different schemes (and otherwise identical parameters) using utilities such as clm dist, clm info on the one hand, and your own sound judgment on the other hand.
If your input graph is extremely dense, with an average node degree (i.e. the number of neighbours per node) that is somewhere above 500, you may need to filter the input graph by removing edges, for example by using one of -tf '#ceilnb()' or -tf '#knn()'.
Shows the explicit settings to which the different preset schemes correspond.
The characteristics are written in the same format (more or less) as the output triggered by -v pruning.
See the -v option below.
These are the different verbosity modes:
To make mcl as quiet as can be, add -q x -V all to the command line.
The -q option governs a general logging mechanism. The format accepted is described in the tingea.log manual page.
The other options govern verbosity levels specific to mcl. -v all turns them all on, -V all turns them all off. -v str and -V str turn on/off the single mode str (for str equal to one of pruning, cls, or explain). Each verbosity mode is given its own entry below.
This mode causes the output of explanatory headers illuminating the output generated with the pruning verbosity mode.
This mode causes output of resource-related quantities. It has a separate entry in the PRUNING OPTIONS section.
This mode (on by default) prints a terse list of characteristics of the clusterings associated with intermediate iterands. The characteristics are E/V, cls, olap, and dd. They respectively stand for the number of outgoing arcs per node (as an average), the number of clusters in the overlapping clustering associated with the iterand, the number of nodes in overlap, and the dag depth associated with the DAG (directed acyclic graph) associated with the iterand. For more information on this DAG refer to the -dump option description in this manual and also mclfaq.
Standard log information
See the -ap option below.
If the -o fname option is not used, mcl will create a file name (for writing output to) that should uniquely characterize the important parameters used in the current invocation of mcl. The default format is out.fname.suf, where out is simply the literal string out, fname is the first argument containing the name of the file (with the graph) to be clustered, and where suf is the suffix encoding a set of parameters (described further below).
The -ap str option specifies a prefix to use rather than out.fname as sketched above. However, mcl will interpret the character '=', if present in str, as a placeholder for the input file name.
If the -aa str option is used, mcl will append str to the suffix suf created by itself. You can use this if you need to encode some extra information in the file name suffix.
The suffix is constructed as follows. The -I f and -scheme parameter are always encoded. Other options, such as -pi f and -knn are only encoded if they are used. Any real argument f is encoded using exactly one trailing digit behind the decimal separator (which itself is not written). The setting -I 3.14 is thus encoded as I31. The -scheme option is encoded using the letter 's', all other options mentioned here are encoded as themselves (stripped of the hyphen). For example
results in the file name out.small.mci.I30s5c25pi08. If you want to know beforehand what file name will be produced, use the -az option.
By default mcl incorporates the input file name into the output file name and appends a short suffix describing the most important option settings. Use -ax to find out what that suffix is. This can be useful in wrapper pipeline scripts such as clxcoarse.
mcl writes the command line with which it was invoked to the output clustering file. Use this option to include any additional information. MCL does nothing with this option except copying it as just described.
Threading is useful if you have a multi-processor system. mcl will spawn k threads of computation. If these are computed in parallel (this depends on the number of CPUs available to the mcl process) it will speed up the process accordingly.
When threading, it is best not to turn on pruning verbosity mode if you are letting mcl run unattended, unless you want to scrutinize its output later. This is because it makes mcl run somewhat slower, although the difference is not dramatic.
If used, mcl will apply inflation one time to the input graph before entering the main process. This can be useful for making the edge weights in a graph either more homogeneous (which may result in less granular clusterings) or more heterogeneous (which may result in more granular clusterings). Homogeneity is achieved for values <num> less than one, heterogeneity for values larger than one. Values to try are normally in the range [2.0,10.0].
The -ph option is special in that it does not rescale columns to be stochastic. Instead, it rescales columns so that the maximum value found in the column stays the same after inflation was applied. There is little significance to this, and what little there is is undocumented.
If used, mcl will apply inflation one time to the input graph before entering the main process. The difference with -pi is that with the latter option mcl may apply certain transformations after reading in the matrix such as adding or modifying loops. The purpose of the -if (mnemonic for inflation-first) option is to use it on graphs saved with the --write-expanded option and convey to mcl that it should not apply those transformations.
Dump during iterations i..j-1. Use all to dump in all iterations. See the -dump str option below.
Sampling rate: select only these iterations in the dump interval. See the -dump str option below.
Set the the stem for file names of dumped objects (default mcl). See the -dump str option below.
str is checked for substring occurrences of the following entries. Repeated use of -dump is also allowed.
lines and cat change the mode of dumping. The first changes the dump format to a line based pairwise format rather than the default mcl matrix format. The second causes all dumped items to be dumped to the default stream used for the output clustering, which is appended at the end.
The ite option writes mcl iterands to file. The cls option writes clusterings associated with mcl iterands to file. These clusters are obtained from a particular directed acyclic graph (abbreviated as DAG) associated with each iterand. The dag option writes that DAG to file. The DAG can optionally be further pruned and then again be interpreted as a clustering using clm imac, and clm imac can also work with the matrices written using the ite option. It should be noted that clusterings associated with intermediate iterands may contain overlap, which is interesting in many applications. For more information refer to mclfaq and the REFERENCES section below.
The result option dumps the usual MCL clustering.
The chr option says, for each iterand I, to output a matrix C with characteristics of I. C has the same number of columns as I. For each column k in C, row entry 0 is the diagonal or 'loop' value of column k in I after expansion and pruning, and before inflation and rescaling. Entry 1 is the loop value after inflation and rescaling. Entry 2 is the center of column k (the sum of its entries squared) computed after expansion and before pruning, entry 3 is the maximum value found in that column at the same time. Entry 4 is the amount of mass kept for that column after pruning.
The -ds option sets the stem for file names of dumped objects (default mcl). The -di and -dm options allow a selection of iterands to be made.
This has two completely different uses. It sets the number of decimals used for pretty-printing mcl iterands when using the --show option (see below), and it sets the number of decimals used for writing the expanded matrix when using the -write-expanded option.
Print matrices to screen. The number of significant digits to be printed can be tuned with -digits n. An 80-column screen allows graphs (matrices) of size up to 12(x12) to be printed with three digits precision (behind the comma), and of size up to 14(x14) with two digits. This can give you an idea of how mcl operates, and what the effect of pruning is. Use e.g. -S 6 for such a small graph and view the MCL matrix iterands with --show.
Write matrix dump output in binary mcl format rather than interchange mcl format (the default). Note that mcxconvert can be used to convert each one into the other. See mcxio and mcx for more information.
str can be one of lex, size, revsize, or none. The default is 'revsize', in which the largest clusters come first. If the mode is 'size', smallest clusters come first, if the mode is 'lex', clusters are ordered lexicographically, and if the mode is 'none', the order is the same as produced by the procedure used by mcl to map matrices onto clusterings.
Mode keep causes mcl to retain overlap should this improbable event occur. In theory, mcl may generate a clustering that contains overlap, although this almost never happens in practice, as it requires some particular type of symmetry to be present in the input graph (not just any symmetry will do). Mathematically speaking, this is a conjecture and not a theorem, but the present author wil eat his shoe if it fails to be true (for marzipan values of shoe). It is easy though to construct an input graph for which certain mcl settings result in overlap - for example a line graph on an odd number of nodes. The default is to excise overlapping parts and introduce them as clusters in their own right. It is possible to allocate nodes in overlap to the first cluster in which they occur (i.e. rather arbitrarily), corresponding with mode cut.
In mode split mcl will put all nodes in overlap into separate clusters. These clusters are chosen such that two nodes are put in the same new cluster if and only if they always occur paired in the clusters of the overlapping clustering.
This option has no effect on the clusterings that are output when using -dump cls - the default for those is that overlap is not touched, and this default can not yet be overridden.
If the input graph has strong bipartite characteristics, mcl may yield clusters that do not correspond to connected components in the input graph. Turn one of these modes on to analyze the resultant clustering.
If loose clusters are found they will be split into subclusters corresponding to connected components. With --force-connected=y mcl will write the corrected clustering to the normal output file, and the old clustering to the same file with suffix orig. With --check-connected=y mcl will write the loose clustering to the normal output file, and the corrected clustering to the same file with suffix coco.
These options are not on by default, as the analysis is currently (overly) time-consuming and mcl's behaviour actually makes some sense (when taking bipartite characteristics into account).
With this mode turned on, mcl will reread the input matrix and compute a few performance criteria and attach them to the output file. Off by default.
Shows the log with process characteristics on STDOUT. By default, this mode is off.
Explains how the jury synopsis is computed from the jury marks.
<int> is interpreted as the number of nodes of an input graph. mcl will print the maximum amount of RAM it needs for its computations. The formula for this number in bytes is:
This estimate will usually be too pessimistic. It does assume though that the average node degree of the input graph does not exceed k. The -how-much-ram option takes other command-line arguments into account (such as -S and -R), and it expresses the amount of RAM in megabyte units.
Shows a selection of the most important mcl options.
Gives a one-line description for all options.
Show current settings for tunable parameters. --show-settings is a synonym.
mcl proceeds as follows. First, entries that are smaller than cutoff are removed, resulting in a vector with at most 1/cutoff entries. The cutoff can be supplied either by -p, or as the inverse value by -P. The latter is more intuitive, if your intuition is like mine (P stands for precision or pruning). The cutoff just described is rigid; it is the same for all vectors. The --adapt option causes the computation of a cutoff that depends on a vector's homogeneity properties, and this option may or may not speed up mcl.
Second, if the remaining stochastic mass (i.e. the sum of all remaining entries) is less than <pct>/100 and the number of remaining entries is less than <r> (as specified by the -R flag), mcl will try to regain ground by recovering the largest discarded entries. The total number of entries is not allowed to grow larger than <r>. If recovery was not necessary, mcl tries to prune the vector further down to at most s entries (if applicable), as specified by the -S flag. If this results in a vector that satisfies the recovery condition then recovery is attempted, exactly as described above. The latter will not occur of course if <r> <= <s>.
The default setting is something like -P 4000 -S 500 -R 600. Check the -z flag to be sure. There is a set of precomposed settings, which can be triggered with the -scheme k option. k=4 is the default scheme; higher values for k result in costlier and more accurate computations (vice versa for lower, cheaper, and less accurate). The schemes are listed using the --show-schemes option. It is advisable to use the -scheme option only in interactive mode, and to use the explicit expressions when doing batch processing. The reason is that there is no guarantee whatsoever that the schemes will not change between different releases. This is because the scheme options should reflect good general purpose settings, and it may become appararent that other schemes are better.
Note that 'less accurate' or 'more accurate' computations may still generate the same output clusterings. Use clm dist to compare output clusterings for different resource parameters. Refer to clm dist for a discussion of this issue.
-warn-pct takes an integer between 0 and 100 as parameter, -warn-factor takes a real positive number. They default to something like 30 and 50.0. If you want to see less warnings, decrease warn-pct and increase warn-factor. Set warn-factor to zero if you want no warnings.
This value (by default set to 10) determines when mcl switches to sparse matrix/vector multiplication. For a given column stochastic vector (corresponding with all the neighbours of a given node v according to the current mcl iterand) the sum S of neighbour counts of all neighbours of v is computed, counting duplicates. This is exactly the number of matrix entries involved in the computation of the new column vector for the matrix product. If S times <int> does not exceed the number of nodes in the graph (equal to both column and row dimension of the matrices used) then a sparse implementation is used. Otherwise an optimized regular implementation is used. Intuitively, this option can be thought of as the estimated overhead per matrix floating point operation incurred by the sparse implementation compared with the regular implementation. MCL uses this estimated overhead to determine which implementation is likely to be quicker. Testing has shown this strategy works very well for graphs of a wide range of sizes, including graphs with up to 3 million nodes and 500 million edges.
The effectiveness of this option is influenced by hardware-specific properties such as the CPU L2 cache size. The default value should work reasonably well across a wide variety of scenarios, but it may be possible to squeeze faster run times out of mcl by tuning this parameter to the graphs that are specific for your application domain.
The following is an example of label input
It can be clustered like this:
The file out.cathat should now like like this
A few things to note. First, MCL will symmetrize any arrow it finds. If it sees bat cat 1.0 it will act as if it also saw cat bat 1.0. You can explicitly specify cat bat 1.0, mcl will in the first parse stage simply end up with duplicate entries. Second, MCL deduplicates repeated edges by taking the one with the maximum value. So,
Will result in two arrows cat-hat and hat-cat both with value 0.8.
mcl will work very well for graphs in which the diameter of the natural clusters is not too large. The presence of many edges between different clusters is not problematic; as long as there is cluster structure, mcl will find it. It is less likely to work well for graphs with clusters (inducing subgraphs) of large diameter, e.g. grid-like graphs derived from Euclidean data. So mcl in its canonical form is certainly not fit for boundary detection or image segmentation. I experimented with a modified mcl and boundary detection in the thesis pointed to below (see REFERENCES). This was fun and not entirely unsuccesful, but not something to be pursued further.
mcl likes undirected input graphs best, and it really dislikes graphs with node pairs (i,j) for which an arc going from i to j is present and the counter-arc from j to i is absent. Try to make your input graph undirected. Furthermore, mcl interprets edge weights in graphs as similarities. If you are used to working with dissimilarities, you will have to convert those to similarities using some conversion formula. The most important thing is that you feel confident that the similarities are reasonable, i.e. if X is similar to Y with weight 2, and X is similar to Z with weight 200, then this should mean that the similarity of Y (to X) is neglectible compared with the similarity of Z (to X).
mcl is probably not suited for clustering tree graphs. This is because mcl works best if there are multiple paths between different nodes in the natural clusters, but in tree graphs there is only one path between any pair of nodes. Trees are too sparse a structure for mcl to work on.
mcl may well be suited for clustering lattices. It will depend on the density characteristics of the lattice, and the conditions for success are the same as those for clustering graphs in general: The diameter of the natural clusters should not be too large. NOTE when clustering a lattice, you have to cluster the underlying undirected graph, and not the directed graph that represents the lattice itself. The reason is that one has to allow mcl (or any other cluster algorithm) to 'look back in time', so to speak. Clustering and directionality bite each other (long discussion omitted).
mcl has a worst-case time complexity O(N*k^2), where N is the number of nodes in the graph, and k is the maximum number of neighbours tracked during computations. k depends on the -P and -S options. If the -S option is used (which is the default setting) then k equals the value corresponding with this option. Typical values for k are in the range 500..1000. The average case is much better than the worst case though, as cluster structure itself has the effect of helping mcl's pruning schemes, certainly if the diameter of natural clusters is not large.
There are currently no resource nor configuration files. The mcl matrix format is described in the mcxio section.
When writing matrices in interchange format, mcl will use this variable (if present) as the precision (number of digits) for printing the fractional part of values.
MCL and its sibling applications will usually report about matrix input/output from/to disk. The verbosity level can be regulated via MCLXIOVERBOSITY. These are the levels it can currently be set to.
MCL and its sibling applications will by default output matrices in interchange format rather than binary format (cf. mcxio). The desired format can be controlled via the variable MCLXIOFORMAT. These are the levels it can currently be set to.
If matrices are output in interchange format, by default empty vectors will not be listed. Equivalently (during input time), vectors for which no listing is present are understood to be empty - note that the presence of a vector is established using the domain information found in the header part. It is possible to enforce listing of empty vectors by setting bit '1' in the variable MCLXICFLAGS.
MCL and its sibling applications will always check a matrix for consistency while it is being read. If this variable is set, the consistency check is omitted. For large graphs the speed up can be considerable. However, if the input graph is not conforming it will likely crash the application that is using it.
If mcl issues a diagnostic error, it will most likely be because the input matrix could not be parsed succesfully. mcl tries to be helpful in describing the kind of parse error. The mcl matrix format is described in the mcxio section.
No known bugs at this time.
Stijn van Dongen.
The MCL algorithm was conceived in spring 1996 by the present author. The first implementation of the MCL algorithm followed that spring and summer. It was written in Perl and proved the viability of the algorithm. The implementation described here began its life in autumn 1997. The first versions of the vital matrix library were designed jointly by Stijn van Dongen and Annius Groenink in the period Oktober 1997 - May 1999. The efficient matrix-vector multiplication routine was written by Annius. This routine is without significant changes still one of the cornerstones of this MCL implementation.
Since May 1999 all MCL libraries have seen much development and redesign by the present author. Matrix-matrix multiplication has been rewritten several times to take full advantage of the sparseness properties of the stochastic matrices brought forth by the MCL algorithm. This mostly concerns the issue of pruning — removal of small elements in a stochastic column in order to keep matrices sparse.
Very instructive was that around April 2001 Rob Koopman pointed out that selecting the k largest elements out of a collection of n is best done using a min-heap. This was the key to the second major rewrite (now counting three) of the MCL pruning schemes, resulting in much faster code, generally producing a more accurate computation of the MCL process.
In May 2001 Anton Enright initiated the parallellization of the mcl code and threaded inflation. From this example, Stijn threaded expansion. This was great, as the MCL data structures and operands (normal matrix multiplication and Hadamard multiplication) just beg for parallellization.
Onwards. The January 2003 03-010 release introduced support for sparsely enumerated (i.e. indices need not be sequential) graphs and matrices, the result of a major overhaul of the matrix library and most higher layers. Conceptually, the library now sees matrices as infinite quadrants of which only finite subsections happen to have nonzero entries.
The June 2003 03-154 release introduced unix-type pipelines for clustering, including the BLAST parser mcxdeblast and the mclblastline script. The April 2004 04-105 release revived binary format, which has been a first class citizen every since.
With the March 2005 05-090 release mcxsubs finally acquired a sane specification syntax. The November 2005 05-314 release brought the ability to stream label input directly into mcl. The subsequent release introduced a transformation language shared by various mcl siblings that allows arbitrary progressions of transformations to be applied to either stream values or matrix values.
Joost van Baal set up the mcl CVS tree and packaged mcl for Debian GNU/Linux. He completely autotooled the sources, so much so that at first I found it hard to find them back amidst bootstrap, aclocal.m4, depcomp, and other beauties.
Jan van der Steen shared his elegant mempool code. Philip Lijnzaad gave useful comments. Philip, Shawn Hoon, Abel Ureta-Vidal, and Martin Mokrejs sent helpful bug reports.
Abel Ureta-Vidal and Dinakarpandian Deendayal commented on and contributed to mcxdeblast and mcxassemble.
Tim Hughes contributed several good bug reports for mcxassemble, mcxdeblast and zoem (a workhorse for clm format).
mclfaq - Frequently Asked Questions.
mcxio - a description of the mcl matrix format.
There are many more utilities. Consult mclfamily for an overview of and links to all the documentation and the utilities in the mcl family.
minimcl is a 200-line perl implementation of mcl. It is shipped in the mcl distribution and can be found online at http://micans.org/mcl.
mcl's home at http://micans.org/mcl/.
Stijn van Dongen, Graph Clustering by Flow Simulation.
PhD thesis, University of Utrecht, May 2000.
 Stijn van Dongen, Graph Clustering Via a Discrete Uncoupling Process, SIAM Journal on Matrix Analysis and Applications, 30(1):121-141, 2008. http://link.aip.org/link/?SJMAEL/30/121/1
Stijn van Dongen. A cluster algorithm for graphs.
Technical Report INS-R0010, National Research Institute for Mathematics and
Computer Science in the Netherlands, Amsterdam, May 2000.
Stijn van Dongen. A stochastic uncoupling process for graphs.
Technical Report INS-R0011, National Research Institute for Mathematics and
Computer Science in the Netherlands, Amsterdam, May 2000.
Stijn van Dongen. Performance criteria for graph clustering and Markov
cluster experiments. Technical Report INS-R0012, National Research
Institute for Mathematics and Computer Science in the Netherlands,
Amsterdam, May 2000.
 Enright A.J., Van Dongen S., Ouzounis C.A. An efficient algorithm for large-scale detection of protein families, Nucleic Acids Research 30(7):1575-1584 (2002).
This page was generated from ZOEM manual macros, http://micans.org/zoem. Both html and roff pages can be created from the same source without having to bother with all the usual conversion problems, while keeping some level of sophistication in the typesetting. This is the PostScript derived from the zoem troff output.