Because of its shared-memory design, PSC’s Blacklight has become an invaluable tool in assembly and analysis of next-generation sequencing data, including projects in behavioral genomics and metagenomics

In genomics, the next generation is now. This relatively new branch of the life sciences has in the last few years, due to new technologies, exploded with possibility and data. “Next generation” sequencing tools have taken genomics well beyond the Human Genome Project to studies of nearly every kind of organism, from ants and bumblebees to Patagonian tuco-tucos (more on that below) among many others, by deciphering the order of nucleotide bases — A, G, C and T (adenine, guanine, cytosine and thymine) — at unprecedented speed.

The essential difference is long versus short reads. Previous sequencers did reads of about 300 to 500 and sometimes up to 1000 bases. The new technologies gain their advantage by doing much shorter reads, 50 to 150 bases — at greatly reduced cost per base — and can generate in a week as much sequence data as would require a year for the traditional sequencers. Consequently, genomics has shifted into data-intensive overdrive, with many opportunities to do important research. While it’s an unprecedented blessing for the life sciences, it’s also an unprecedented challenge for data processing and analysis.

PHOTO:Phil Blood

Phil Blood, XSEDE Extended Collaborative Support Services consultant, PSC

Once a sequencing instrument has produced millions or, as the case may be, billions of reads from an organism’s DNA (or RNA), researchers face the task of assembling them. To add to the degree of difficulty, short reads amplify the computational challenge — many more pieces of data must be fit together based on shorter overlaps. How do you assemble all those sequence fragments into complete and accurate genomic strands? Imagine a jigsaw picture puzzle with 100 big pieces versus the same picture with 2,000 little pieces. It’s a potentially mind-boggling problem handled by very sophisticated computational algorithms, requiring many runs, careful checking and, some say, as much art as science.

Since Blacklight came online in October 2010, with two partitions of 16 terabytes of shared memory, the largest shared-memory system in the world, it’s become a powerful tool in meeting this challenge. An article in GenomeWeb (February 1, 2012) highlighted Blacklight’s advantages — very large shared memory making it possible to contain entire base-pair datasets in random-access memory (RAM) — dramatically improving workflow and throughput time, as compared to non-shared memory clusters, for genomics assembly and analysis.

Beyond that, observes PSC scientist and XSEDE consultant Phil Blood, many large genomics assemblies simply couldn’t be done without large shared-memory, such as Blacklight provides. To enhance Blacklight’s genomics capabilities, Blood has made nearly all genomics software tools available for easy use — saving much time for researchers. “Blacklight has an extensive collection of pre-compiled modules for the analyses of next-generation sequence data,” says Matthew MacManes, of the University of California, Berkeley. MacManes attempted genomics analysis with a number of high-end computing systems before coming to Blacklight. He used nearly 20 different programs in his assembly and analysis at PSC: “Having these programs installed and maintained by PSC staff is extremely helpful.”

MacManes’ project with Blacklight involved assembly and analysis of RNA from tuco-tucos, a burrowing rodent from Patagonia — of particular interest in that some tucos live in social groups while others of the same species are anti-social and live alone. MacManes identified a number of genes that are expressed or not depending on whether the tucos live alone or in a colony. “For my research,” says MacManes, “the use of Blacklight has been absolutely revolutionary, allowing me to complete analyses that link specific patterns of gene expression with mammalian social behavior.”

In science, new tools often make it possible to look at new questions, and the availability of next-generation sequencing has led to studies in “metagenomics” — analysis of genes from many organisms that co-exist in the same place — unimaginable a few years ago. “In metagenomics, unlike the traditional approach of analyzing the genome of one organism, you can pick an environment and take a sample,” says Blood. “It could be Old Faithful, or what’s inside human intestines, wherever you might find interesting microbial communities.”

In a metagenomics study with Blacklight, Blood helped researchers from Oklahoma State assemble sequencing data from soil that came from a sugar-cane plantation in Brazil. The goal is to find enzymes that can efficiently break down non-feed plants, such as switchgrass, wheat straw and others that have the potential to yield biofuel more efficiently than feed-stock plants like corn. Thanks to Blacklight, the Oklahoma State team — Mostafa Elshahed, Rolf Prade and Brian Couger, with Couger handling the computation — completed the largest metagenomics assembly to date.

“It wouldn’t have been possible for us to do this on any other system,” says Couger. Their work, still in analysis, has identified thousands of candidate enzymes, all previously unknown, that offer promise to cost-effectively degrade non-feed-stock crops to biofuel.

Behavioral Genomics: Alone Time for Tucos

PHOTO:Phil Blood

Matthew MacManes

“Does behavior evolve through gene expression changes in the brain in response to environment?” The question, posed by a leading genomics scientist, Gene Robinson, caught MacManes’ attention in 2009, when he was challenged by his laboratory group leader, Eileen Lacey, to come up with a dream research project. The answer to Robinson’s question, says MacManes, is “yes, but . . . which genes, and how do we find them?”

Evolutionary biology explains that species maintain group behavior when the survival benefit is greater than cost — birds flock, for example, wolves run in packs, and tigers are territorial and mainly solitary. With these and many other examples, how do differences in social behavior show up in genes? Studies have identified a few genes that appear to be involved across a number of unrelated species, mostly insects, but there’s little consensus, says MacManes, about the genetic underpinnings of social versus solitary behavior.

These thoughts led MacManes to what he calls his “craziest idea ever” — which now, only three years later, because of rapidly evolving genomics technologies, seems not so crazy at all. What if, he thought, he could look at the genomics in a case where both ends of the solitary versus social spectrum are represented in the same species?

The colonial tuco-tuco (Ctenomys sociabilis) in its native environment (volcanic ash). A burrowing rodent, tucos are of interest in behavioral genomics because they vary, in the same species, between social and solitary living conditions.

Conveniently, such a species, the colonial tuco-tuco (Ctenomys sociabilis) was available, a population of them housed and studied at Berkeley’s Museum of Vertebrate Zoology, where MacManes is an NIH-sponsored post-doctoral fellow. The colonial tuco-tuco — so-called for a clicking sound it makes — is a subterranean, burrowing rodent from Patagonia, related to the common guinea pig, unusual in the intra-species variation in behavior it exhibits. “Some of the females,” says MacManes, “live in colonies with larger family groups, while others — at about one year of age — disperse from their birth burrow and live alone. Most social animals are obligately social — there aren’t usually solitary animals to be found, and this variation makes tucos interesting and unique.”

Working with two control populations of five tucos each, housed in social and solitary conditions, MacManes used messenger RNA from the hippocampus — a brain region implicated by prior research in social behavior. The extracted tissue was sequenced (in an Illumina sequencer), yielding 56 billion base-pairs of raw data — 560 million 100 base-pair reads, “a ridiculous amount of reads,” says MacManes.

To grapple with assembly and analysis of this huge amount of sequence data, beginning with the task of building complete “transcriptomes” — full strands of RNA — MacManes first turned to large distributed-memory machines at several sites, but eventually came to PSC’s Blacklight. Using 80 cores of Blacklight, 640 gigabytes of RAM, he completed the assembly in 14 days of computing, with subsequent analysis extending for months.

“Currently there is simply no better resource for this type of work.”

The work identified a number of genes that are differentially expressed depending on tuco social behavior, and MacManes and Lacey have a manuscript in preparation reporting their findings. “Blacklight is a key resource for my analyses of next-generation sequence data,” says MacManes. “Without it, I would simply have been unable to complete the requisite analyses. I feel so strongly about Blacklight that I have referred colleagues and collaborators. Currently there is simply no better resource for this type of work.”

Metagenomics: Looking for Biofuel Enzymes

PHOTO:Phil Blood

Brian Couger (right), Rolf Prade (center) and Tyler Weirick, Oklahoma State University

“In Brazil,” says Rolf Prade, a professor of microbiology at Oklahoma State University and a Brazilian, “biofuel is standard gasoline for cars.” The world’s second largest producer of biofuel, Brazil gets its ethanol from sugarcane, uniquely available there due to enormous amounts of arable land and suitable climate.

Like corn in the United States, however, sugarcane has major disadvantages as a biofuel source. The amount of energy available per amount of input crop is much greater in non-feed stock plants, with denser fibrous structure (lignocellulosic plants), such as switchgrass and wheat straw, but these non-food plants are expensive and difficult to degrade into biofuel.

Considerable research worldwide is focused on finding inexpensive means to overcome this obstacle. A team at Oklahoma State led by Mostafa Elshahed, along with Prade and Brian Couger, has taken an innovative metagenomics approach. To find enzymes that can do the heavy-duty biodegrading, they started with a soil sample, gathered by Prade, from a Brazilian sugar-cane field.

The 1979 Brazilian Fiat 147 was the first modern automobile capable of running only on ethanol, which Brazil produces from sugarcane, shown ready for harvest at a plantation in Sao Paulo State.

“These fields have been growing for 50 years,” says Prade. “They cut the stems off the plants and throw everything else back in the soil and let it recycle — so this is very efficient recycling soil.” To further accumulate microorganisms involved in biomass degradation, the researchers enriched the sample in a bioreactor with oxygen and more sugarcane biomass, and cultured the soil for eight weeks.

To isolate DNA, the researchers then sequenced the soil (in an Illumina next-generation sequencer). This yielded 1.5 billion pair-data reads of 100 basepairs each, approximately 300 gigabases in total. With Couger handling the computing, the researchers turned to Blacklight for the assembly, using software called Velvet — made available on Blacklight as a pre-compiled module by XSEDE consultant Blood of PSC.

The entire metagenomics dataset occupied 3.5 terabytes of Blacklight memory. “This is the largest metagenomics assembly ever done,” says Couger, “and it would have been intractable on any computational cluster other than Blacklight.” To make use of this assembled data as a means to identify enzymes, the researchers also did a series of protein separations from the soil samples and applied mass spectrometry. From precise molecular weights, they inferred amino-acid sequences. Matching these sequences with the assembly data, the researchers identified more than 8,000 gene candidates related to glycoside hydrolase, a category of enzymes that can degrade plant cell walls.

“This is the largest metagenomics assembly ever done, and it would have been intractable on any computational cluster other than Blacklight.”

“This is like a protein discovery platform,” says Prade. “Making biofuels from lignocellulosics doesn’t work because we don’t know how to decompose the biomass. It’s because we don’t have all the proteins, and we’re working to find those.”