khmer v2.0 release notes

This is the v2.0 release of khmer and the first from our new lab at the University of California, Davis. It features Python 3 compatibility, streaming I/O from Unix Pipes, mixed-pair sequence file format support, and a new parameter to simplify memory usage. We also have a software paper in-press describing the project and the citation reminders have been updated to reflect that.

Overall there are an additional 2,380 lines of Python code (mostly tests) and 283 less lines of C++ (despite adding features). This release is the product of over 1,000 commits to the codebase since v1.4.

Documentation is at

New items of note:

New behavior

Streaming I/O from Unix Pipes

All scripts now accept input from named (like /dev/stdin, or that created using <( list ) process substituion) and unnamed pipes (like output piped in from another program with |). The STDIN stream can also be specified using a single dash: -. #1186 @mr-c #1042 #763 @SherineAwad #1085 @ctb

New parameter for memory usage, and/or tablesize/number of table parameters.

There is now a -M/--max-memory-usage parameter that sets the number of tables (-N/--n\_tables) and tablesize (-x/--max-tablesize) parameters automatically to match the desired memory usage. #1106 #621 #1126 #390 #1117 #1055 #1050 #1214 #1179 #1133 #1145 @ctb @qingpeng @bocajnotnef

Digital normalization script now supports mixed paired and unpaired read input now supports mixed paired and unpaired (or "broken-paired") input. Behavior can be forced to either treat all reads as singletons or to require all reads be properly paired using --force\_single or --paired, respectively. If --paired is set, --unpaired-reads can be used to include a file of unpaired reads. The unpaired reads will be examined after all of the other sequence files. now has a --quiet option to reduce the amount of output. #1200 @bocajnotnef

Mixed-pair sequence file format support --output-orphaned/-0 has been added to allow for orphaned reads and give them a file to be sorted into. #847 #1164 @ctb

Scripts now output columnar data in CSV format by default

All scripts that output any kind of columnar data now do so in CSV format, with headers. Previously this had to be enabled with --csv. (Affects,,, and --report also now outputs in CSV format. #1011 #1180 @ctb

Reservoir sampling script extracts paired reads by default now retains pairs in the output, by default. This can be overridden to match previous behavior with --force\_single.

Most input and output files can be compressed

We support gzip and bzip2 input and output file compression everywhere that it makes sense #505 #747 @bocajnotnef

New scripts

Estimate number of unique kmers estimates the k-mer cardinality of a dataset using the HyperLogLog probabilistic data structure. This allows very low memory consumption, which can be configured through an expected error rate. Even with low error rate (and higher memory consumption), it is still much more efficient than exact counting and alternative methods. It supports multicore processing (using OpenMP) and streaming, and so can be used in conjunction with other scripts (like and This script is the work of @luizirber and the subject of a paper in draft. #390 #1239 #1252 #1053 #1072 #1145 #1176 #1207 #1204 #1245

Incompatible changes

New datastructure and script names

For clarity the Count-Min Sketch based data structure previously known as "counting_hash" or "counting_table" and variations of these is now known as countgraph. Likewise with the Bloom Filter based data structure previously known at "hashbits", "presence_table" and variations of these is now known as nodegraph. Many options relating to table have been changed to graph. #1112 #1209 @mr-c

Binary file formats have changed

All binary khmer formats (presence tables, counting tables, tag sets, stop tags, and partition subsets) have changed. Files are now pre-pended with the string OXLI to indicate that they are from this project. #519 #1031 @mr-c #1159 @luizirber

Files of the above types made in previous versions of khmer are not compatible with v2.0; the reverse is also true.

In addition to the OXLI string, the Nodegraph and Countgraph file format now includes the number of occupied bins. See for details. #1093 @ctb @mr-c #1101 #1103 @kdmurray91 no longer appends .pt to the specified filename

Previously, appended a .pt extension to the specified output filename and appended a .pt to the given input filename. Now, writes to the specified output filename and does not append a .pt to the given input filename. #1189 #747 @bocajnotnef

Some reporting options have been turned always on

The total number of unique k-mers will always be reported every time a new countgraph is made. The --report-total-kmers option has been removed from,, and to reflect this. Likewise with --write-fp-rate for and; the false positive rate will always be written to the .info files. #1097 #1180 @ctb

An uncommon error recovery routine was removed

To simplify the codebase --save-on-failure and its helper option --dump-frequency have been removed from

Single file output option names have been normalized

--out is now --output for both and #1188 #1164 @ctb

Miscellaneous changes

The common option --min-tablesize was renamed to --max-tablesize to reflect this more desirable behavior.

In conjuction with the new --output-orphaned option, the option --force-paired/-p has been eliminated.

As CSV format is now the default, the --csv option has been removed.

Removed script has been removed.

Notable bugs fixed/issues closed:

When decides to keep both parts of a pair of reads it was only adding the k-mers & counts from one to the countgraph. #1000 #1010 @drtamermansour @bocajnotnef

The partition map file format was not robust to truncation and would hang waiting for more data. #437 #1037 #1048 @ctb and no longer create default files when the user supplies filename(s). #1005 #1132 @kdmurray91

Additional fixes/features was missing a --force option and unit tests. #358 #1078 @ctb The check for excessively high false-positive rate has also received a --force option #1168 @bocajnotnef

A bug leading to an infinite loop with large gzipped countgraphs was found #1038 #1043 @kdmurray91

All scripts that create nodegraphs or countgraphs report the total number of unique k-mers. #491 #609 #429 @mr-c

Read pairs from SRA are fully supported. Reported by @macmanes in #1027, fixed by @kdmurray91 @SherineAwad in #1173 #1088

Of interest to users:

Added Hashtable::get_kmers(), get_kmer_hashes(), and get_kmer_counts() with corresponding CPython functions. #1047 #1049 @ctb

The DEFAULT_DESIRED_COVERAGE for is now 20. #1073 #1081 @ctb

FIFOs are no longer seen as empty. #1147 #1163 @bocajnotnef

When the k-size is requested to be larger than 32 (which is unsupported) a helpful error message is reported. #1094 #1050 @ctb

We try to report more helpfully during errors, such as suggesting the --force option when outputs files already exist. #1162 #1170 @bocajnotnef

There is a paper related to "Crossing the streams: a framework for streaming analysis of short DNA sequencing reads" and it has been added to the CITATION file and program output. #1180 #1130 @ctb

We have dropped support for Python 2.6 #1009 #1180 @ctb

Our user documentation got a bit out of date and has been updated. #1156 #1247 @bocajnotnef @mr-c #1104 @kdmurray91 #1267 @ctb Links to lists of publications that use khmer have been added #1063 #1222 @mr-c The help text from the scripts has also had a thorough cleanup for formatting. #1268 @mr-c's --n_keep option has incorrect help text. We now point out that all reads with Ns will be dropped by default unless this option is supplied. #657 #814 #1208 @ACharbonneau @bocajnotnef

We've updated the URL to the '88m-reads.fa.gz' file. #1242 #1269 @mr-c

@camillescott designed and implemented an optimization for #862 can now be used without counts over 255 with --no-bigcount. #1067 #909 @drtamermansour @bocajnotnef Its input file requirement can no longer be overridden #1201 #1202 @bocajnotnef

khmer v2.0 will be released as a package for the Debian GNU/Linux operating system. Big thanks to @kdmurray91 for his assistance. #1148 #1240 The C++ library, now named liboxli, will have its own package as well.

sandbox/ now wraps long FASTA sequences at 80 columns. #450 #1136 @SherineAwad

Of interest to developers:

The khmer project is now a Python 3 codebase with backwards compatibility to Python 2.7. Huge credit to @luizirber #978 #922 #1045 #1066 #1089 #1157 #1191 #1108 Many developer impacting changes including the file khmer/\ is now khmer/\ #169 #904

@camillescott did an extensive refactor of the C++ graph traversal code which removed a considerable amount of redundant code and will be very useful for future work. #1231 #1080

We now use some and allow all C++11 features in the codebase. #598 #1122 @mr-c was extensively refactored. #1006 #1010 #1057 #1039 #1135 #1182 @bocajnotnef @ctb @camillescott

The CPython glue was refactored so that CountingHash and Hashbits inherit from Hashtable. #1044 @ctb

The tests no longer stop on the first failed test. #1124 #1134 @ctb and some noisy tests were silenced #1125 #1137 @bocajnotnef

The check_space() calls were cleaned up. #1167 #1166 #1170 #993

Developer docs have been expanded #737 #1184 @bocajnotnef #1083 #1282 @ctb @mr-c #1269

A lot of code was deleted: TRACE related code in #274 #1180 @ctb hashtable_collect_high_abundance_kmers in #1142 #1044 @ctb lib/, lib/, lib/ #1144, @mr-c bink.ipynb, lib/, lib/primes.hh #1289 @mr-c

@bocajnotnef deleted more unused code and added new tests elsewhere to increase testing coverage in #1236. @mr-c had his own go in #1279

cppcheck installation for OSX has been documented #777 #952 #945 @elmbeech

ccache and git-merge-changelog has been documented for Linux users #610 #1122 #614 @mr-c

The graphalign parameters can be saved/loaded from disk. In addition the align_forward method has been introduced. #755 #750 @mr-c @ctb

labelhash is now known as graphlabels #1032 #1209 @mr-c It is also now a 'friend' of Hashtable and one can make either a nodegraph or countgraph version. These graphlabels can now be saved & loaded from disk. #1021 @ctb

Spelling is hard; we've added instructions on how to run codespell to the developer docs. #890 #1203 @bocajnotnef

A redundant and contradictory named test has been removed. Reported by @jgluck in #662 fixed by @bocajnotnef in #1220 @SherineAwad contributed some additional tests #809 #615.

The new oxli command, while disabled in the v2.0 release, has been added to all the QA makefile targets as we continue to refactor the codebase. #1199 #1218 @bocajnotnef

The CPython code was audited to ensure that all possible C++ exceptions were caught and dealt with. The exception hierarchy was also simplified #1016 #1015 #1017 #1151 @kdmurray91 @mr-c

get_kadian_count has been removed. #1034 #1194 @ctb

We use argparse's metavars to aid with autogenerated documentation for the scripts. This has been documented in the dev docs. #620 #1222 @mr-c

Sometimes one makes a lot of commits while refining a feature or pull request. We've documented a field-tested way to turn a pile of commits into a single commit without the pain of git rebase. #1013 #660 #1222 @mr-c

We use Coverity to test for various issues with our C++ code. The Makefile target has been updated for changes on their side. #1007 #1222 @mr-c

There is a new update() function to merge two nodegraphs of the same size and ksize. #1051 @ctb

Despite the checklist, formatting errors still occur. We must be vigilant! #1075 @luizirber

There is a new filter_on_median function. #862 #1077 @camillescott

There are new scripts in the sandbox/ which output k-mer counts: sandbox/{,}. #983 @ctb

A large effort to make the codebase 'pylint clean' has begun with #1175 @bocajnotnef Likewise the cpychecker tool was re-run on the CPython code and issues found there were addressed #1196 @mr-c

As repeatedly promised, we've updated our list of contributors to include everyone with a commit in git. #1023 @mr-c

thread_utils.is_pair() has been dropped in favor of utils.check_is_pair() #1284 @mr-c

The Doxygen produced documentation is improving. The location of included headers is now autodetected for Doxygen and cppcheck.

Known issues: in multithreaded mode will find slightly different number of unique kmers. This is being investigated in #1248


@ctb, @bocajnotnef, @mr-c, @luizirber, @kdmurray91, @SherineAwad, @camillescott, ‡@ACharbonneau

‡ Indicates new contributors

Issue reporters

@jgluck, @ACharbonneau, @macmanes

comments powered by Disqus