July 27, 2014 Leave a comment

ggplot2 sounds like a great package but I currently don’t have the optimism bandwidth to use it :(

I can, to some extent, forgive removal of relative height/width parameter for facet grids, but I strictly maintain that axis positioning (top/bottom, left/right) must be a basic feature of any decent plotting package.

Categories: Uncategorized

Getting Bibtex entries from DOI (command line)

May 12, 2014 1 comment

Here’s how:

#!/usr/bin/env bash
curl -LH “Accept: text/bibliography; style=bibtex”$1

Categories: Uncategorized Tags:

Performance of Sambamba MRI bindings

October 27, 2013 Leave a comment

So I hacked up a simple BAM reader yesterday as a C extension. For now, BamRead class in Ruby only exposes a ‘name’ method just for the sake of not being totally useless. After all, what I’m interested in is performance of conversion from D to Ruby. And with careful coding the cost is not that large. (What I love about C extensions is full control over what happens.)

In fact, on multicore systems it easily beats PySAM (this simple test is counting reads and computing average read name length; the file contains Ion Torrent 200bp data, 570MB in size)

Python 2.7.3 + PySAM
$ time python

real    0m13.034s
user    0m12.588s
sys     0m0.412s

Ruby 2.1.0-preview1
$ time LD_LIBRARY_PATH=. ruby test.rb

real    0m4.820s
user    0m16.756s
sys     0m0.700s

However, the dynamic loading of my D library sometimes just hangs due to some deadlock =\ I currently use 2.063 where support for shared libraries is not official, because in 2.064 there’s an issue with zlib that I reported to their bugtracker. Hopefully that will be resolved soon.

Categories: Uncategorized

Ruby: FFI or not FFI?

October 26, 2013 Leave a comment

I started making Ruby bindings for my SAM/BAM library, and it’s not at all clear whether to use FFI or good old C extension for MRI.

For you to get the clear picture, I’m going support only Linux and Mac OS X, distributing binary packages, because that’s the easiest option given the current state of DMD compiler infrastructure—it’s not available by default on almost every system, like GCC.

One factor is convenience. By that word FFI proponents usually mean that they are too lazy to sit and write some C code. But hey, since I’m going to distribute binary packages only, I can just use Rice which should be much easier.

Another important factor is interoperatibility. Well, I did some simple benchmarks and discovered that in MRI, the speed sucks if I use FFI, and in JRuby, simply calling Picard library gives the same or better performance (of course, if you’re aware of flags –server and -Xji.objectProxyCache=false). But more importantly, on JRuby the overhead with either FFI or Java Integration is huge =\ Namely, counting reads in a 2.5GB file took about 2 minutes, but adding computation of average read name length added another 2 minutes, giving total of 4! For comparison, it took only 2 minutes using PySAM, and this is the baseline that should be followed.

My conclusions from this are that
1) JVM is not well suited for dynamic languages, and this opinion is supported by JRuby developers. Hopefully, Topaz will mature eventually.
2) Overhead of FFI is too substantial to ignore in my particular case, where we want to work with lots of short reads.

So, I will go with Rice. The bonus part is that I will have to write C++ classes wrapping D functionality, which could theoretically be also used for CPython extension using Boost.Python.

EDIT:  compared bindings generated with Rice and SWIG, the latter wins. So, the full chain is D -> C bindings -> C++ wrapper -> SWIG wrapper -> Ruby

EDIT2: after some playing with SWIG I realized that I want ultimate control over what’s going on, and finally decided to write in good old C.

Categories: Uncategorized Tags: , ,

First experience with Mono-D

December 16, 2012 Leave a comment

Today I installed Mono-D, hoping that it might make my experience with D a little better. Unfortunately, it was far from my expectations. When it is used on small pieces of code, everything looks fine. But when I added the whole BioD library (which I slowly develop) to the include paths, performance decreased drastically and it became impossible to type code. Each autocompletion took at least 30 seconds to resolve :-( No thanks, I would rather type method name manually, 10 times faster.

And even neglecting slowness, completion results are also far from what I need. I use a lot of templates and `auto` declarations that seem to confuse Mono-D engine. I also had no luck with type deduction of iteration variable in foreach statements.

Thus, I’m staying with Vim, although it’s totally dumb when it comes to the language semantics. Hopefully, Mono-D will improve over years. Perhaps, it already works well for Java-style projects that don’t use templates all over the place, but that’s not my case.

UPDATE (March 06, 2013)

Tried again with Xamarin Studio and Speed issue has been resolved, but autocompletion still doesn’t play well with templated code.

Categories: Uncategorized

New cool stuff in Sambamba!

October 5, 2012 Leave a comment

During last month, I have introduced quite a few cool features in Sambamba library. This blog post is a brief overview of them.

MD tags -> reference bases

If reads in the BAM file have MD tags (e.g. calculated with the aid of samtools calmd), you can reconstruct parts of reference sequence from them. This is done lazily, without any heap allocations. If you need a string instead of a lazy character sequence, just wrap it with to!string.

Example (taken from

    auto read = Alignment("r4",
                         [CigarOperation(2, 'M'),
                          CigarOperation(1, 'I'),
                          CigarOperation(7, 'M'),
                          CigarOperation(6, 'D'),
                          CigarOperation(26, 'M')]);
    read["MD"] = "3C3T1^GCTCAG26";

The same works for ranges of coordinate-sorted reads aligned to the same reference. That is, in this case no heap-allocations occur as well, and the sequence is also lazy.

Notice that last two reads in the following example don’t overlap, therefore space between them is filled with ‘N’.

(in this example, some bases are skipped to fit into screen)

    auto r1 = Alignment("r1",
                        [CigarOperation(89, 'M'),   CigarOperation(1, 'S')]);
    r1.position = 60246;    r1["MD"] = "89";

    auto r2 = Alignment("r2",
                        [CigarOperation(83, 'M'),   CigarOperation(7, 'S')]);
    r2.position = 60252;    r2["MD"] = "82T0";

    auto r3 = Alignment("r3",
                        [CigarOperation(90, 'M')]);
    r3.position = 60283;    r3["MD"] = "90";

    auto r4 = Alignment("r4",    "CCCTTGCTGAC...TATCCAACCG",   [CigarOperation(90, 'M')]);
    r4.position = 60337;    r4["MD"] = "90";

    auto r5 = Alignment("r5",    "GAGGCTCCACCC...AAGCACTTTGT", [CigarOperation(90, 'M')]);
    r5.position = 60432;    r5["MD"] = "90";

    auto reads = [r1, r2, r3, r4, r5];
    assert(equal(dna(reads), "AGGTTTTGTG...CTATCCAACCGNNNNNGAGGC...CTTTGT"));


This feature requires dcollections to be installed into Sambamba root directory. Installation process is described at the top of pileuprange.d file, this will also be in README soon.

The behaviour might be a bit different from samtools with respect to insertions/deletions, I’ll revisit it later. For now the main focus is SNPs.

Some example code showing how to work with pileup in Sambamba:

foreach (column; pileupWithReferenceBases(reads)) {
    writeln(column.position, ' ', column.coverage);
    foreach (read; column.reads) {
        if (read.current_base != column.reference_base &&
            read.current_base_quality > 30 && read.cigar_operation == 'M')

Improved efficiency of random access code

Today I have revisited code related to random access. A nice idea came to my mind, and I parallelized decompression much better than before.

For instance, here are the times for counting alignments from NA12043.chrom20.LS454.ssaha2.CEU.low_coverage.20101123.bam.v1, overlapping the region 20:1261261-12162614 (the number of such reads is 63417).


real 0m0.579s
user 0m0.560s
sys 0m0.020s

sambamba 0.2.8

real 0m0.349s
user 0m0.616s
sys 0m0.104s

sambamba trunk

real 0m0.152s
user 0m0.648s
sys 0m0.120s

The experiment was done, as usual, on Pjotr’s 8-core machine. These results mean, basically, that level of parallelism for accessing a region using BAI index is now the same as for serial access. I will update the Debian package soon :)

Categories: Uncategorized

GSoC weekly report #12

August 20, 2012 5 comments

This is my final GSoC report. Sambamba improved significantly over the last two weeks. Several bugs were found and fixed, and some new functionality was added.

Pipeline support

Sambamba library can now work with arbitrary streams, including non-seekable ones (of course, random access is out in this case). However, I haven’t yet figured out how to deal with ‘-‘ command-line parameters using std.getopt module, so please use /dev/stdin and /dev/stdout for the time being.

MessagePack output

Seems like earlier I’ve underestimated the performance boost it can bring :)

I’ve measured time of looping through all alignments from a 112MB BAM file, and got the following results (on Pjotr’s 8-core box):

  • MRI Ruby, JSON input with ‘oj’ gem — 26.0s real, 29.2s user
  • MRI Ruby, MsgPack input — 6.7s real, 9.3s user
  • Python, JSON input with ‘ujson’ — 17.8s real, 21.0s user
  • Python, MsgPack input — 3.5s real, 7.3s user

Just for comparison, time of looping through the file with PySAM (fetch() method) is 3.4s real/3.4s user, i.e. in case of MsgPack output multithreaded execution compensates for additional serialization/deserialization steps.

I’m pretty sure output speed can be further improved—I didn’t yet tweak it thoroughly.

Ruby gem update

Now it works with MsgPack output format, also exception handling is improved. It’s not yet on Travis CI, however, and works only with Ruby 1.9 — I use Open3 for checking stderr to throw meaningful exceptions.

I’ve also set up Travis CI hooks. MRI 1.9.2 and 1.9.3 pass tests. However, I can’t get JRuby working there due to the same issues as with BioRuby—something related to popen/popen3. But JRuby in 1.9 mode works fine, trust me :)

Galaxy toolshed wrapper

I’ve managed to somehow make a wrapper for filtering tool:

However, I can’t say that I like Galaxy. Using command-line is way faster than waiting for some slow Python engine to do the same job. Well, I’m just not that kind of person Galaxy was created for :)

Future plans

Next step is making a pileup engine. That’s the most essential part which is lacking in my tools now. For the design I’ll take ideas from PySAM and bio-alignment gem. (Also it involves using statistics for variant calls and thus is related to my studies.) Hopefully, then I’ll get more feedback.

Another direction of further development is making a decent validation tool. I’m sure I can make a better one that ValidateSamFile from Picard (more flexible and faster) but at the moment motivation is lacking.


I’ve learned quite a bit about bioinformatics. This is definitely a very interesting field, and amounts of data to be analyzed are growing rapidly. I won’t be surprised if BAM will be replaced by CRAM or some another format in a few years, and that will render my library useless. Nevertheless, now I’ve got a good experience of writing libraries of that sort, and tweaking the performance. That will surely be of great help in the future.

Thank you Google and Open Bioinformatics Foundation :)

Categories: Uncategorized Tags:

Get every new post delivered to your Inbox.