Happy New Year
I know it is February 9th, over one month past when we traditionally send New Year’s cards in USA. However, we have a global readership and someone in some corner of the world must be celebrating arrival of new year today. :) We were in Singapore last week watching the city adorned with giant dragons on occasion of Chinese New Year. Thailand and many parts of India will begin their new year in about two months. No matter where you are, we wish you happiness and success in the days and months ahead.
We apologize for the brief hiatus of two months from reporting on bioinformatics. Those of you, who emailed us, expect to receive replies shortly. This absence was not due to nothing newsworthy happening in the world of transcriptomics, but we ourselves got very busy developing and testing a new algorithm for genome assembly. Its main purpose was to reduce the amount of memory required to assemble a large genome, and to develop an approach that could be implemented in a cluster in parallel fashion (without inter-process communication as in ABySS). Our new algorithm worked as a proof of concept, but it would need some work before an out-of-the-box software program can be presented to the community. Nevertheless, working on it helped us understand many of the intricacies of short read data sets and assemblies, whereas most users of readily available programs are limited to tweaking parameters blindly. We will follow up on this topic in the coming days.
In the meanwhile, several promising discoveries were reported by various groups to excite our readers. We shall list them here and delve into their details in later posts.
1. Transcriptome assembler from BGI
We have not covered the details of SOAPdenovo assembler from BGI, but it is one of the strongest genome assemblers for next-generation sequences. This assembler uses de Bruijn graph approach just like Velvet, ABySS, Trinity and other available programs. BGI recently reported a transcriptome module ‘SOAPdenovo-Trans’ that can be useful for transcriptome assembly. We have not tested it ourselves, but users of SOAPdenovo may find the module handy.
2. Genome assembler SGA that uses Burrows-Wheeler transform
An interesting paper titled ‘Efficient de novo assembly of large genomes using compressed data structures’ came out of R. Durbin’s group. They used FM-index derived from compressed Burrows-Wheeler transform, and performed assembly of 1.2 billion sequence reads from human genome with only 54GB of memory. Regular readers of our blog know, how use of Burrows Wheeler transform dramatically improved speed of sequence searching from MAQ-like programs to Bowtie/BWA-like program. It was a matter of time, before someone tried the same method on assemblies instead of searches.
3. Sequence-specific error profile of Illumina sequencers
The archives of Velvet and Oases mailing lists have many interesting discussions on how to improve efficiency of assembly runs, because everyone is limited by RAM size and time. Pre-filtering of reads is one common theme echoed by almost every expert, but apart from few reads with Ns and low- complexity sequences, what can one reasonably filter out? An interesting paper came out in Nucleic Acids Research from a Japanese group. It showed that NGS reads from Illumina machines often contain sequence-specific errors near (i) inverted repeats and (ii) GGC sequences. They observed this error profile in all the Illumina sequencing data they examined regardless of the source organisms or the sample preparation methods used.
These kinds of technology-dependent errors often confuse people as we can recall from our experience. Several years back, we were working on tiling array data and saw very strong signals in some unannotated genomic regions. We tested those regions further using Northern blot hoping that they could be novel non-coding RNA, but did not get any positive result. Later on, we compared tiling array data from nine different organism and found out that T7 in vitro transcription system was a major source of artifact. You can read the details here.
4. Trinity - color space
Several of you gave me feedback on the color space modification of Trinity that we developed, and the results are no different from what we experienced and reported earlier. The number of assembled transcripts were too few compared to what one would expect in a comparable organism and what one often gets from nucleotide space transcriptome assemblers. The reason, we suspect, is higher noise level in color space data. This is a topic we plan to investigate further after completing the genome assembly algorithm that we are working on.
