Overview

The Sorghum bicolor genome project was initiated through the DOE-JGI Community Sequencing Program (CSP) by a consortium led by Andy Paterson, John Bowers, Steve Kresovich, C. Thomas Hash, Jo Messing, Daniel Peterson, Jeremy Schmutz, and Dan Rokhsar.

Large-scale shotgun sequencing of sorghum began at the end of 2005 and was completed on January 25th, 2007. A total of 10,717,203 shotgun reads were collected. All raw trace data is deposited in the NCBI Trace Archive in accordance with our commitment to early access and the Fort Lauderdale genome data release policy .

The present v1.0 release, comprising the Sbi1 assembly and Sbi1.4 gene set, are the assembly and annotation used in the sorghum genome paper. In all subsequent releases, chromosome and gene identifiers will be mapped forward whenever possible. This assembly was built with Arachne v20070201 with a data freeze from January 25th, 2007. After the build, 28 breaks were made and 108 manuals joins were performed. Ten of these joins were across centromeres. The size of the centromere was estimated for each chromosome from the amount of centromeric sequence already assembled. The main genome is in 10 chromosomes with many small unmapped pieces, some of which contain annotated genes. coordinate.

The Sorghum bicolor genome has been published and is available from Nature:

Paterson AH, et al. (2009). "The Sorghum bicolor genome and the diversification of grasses." Nature 457, 551-556 (29 January 2009) | doi:10.1038/nature07723

Statistics

Genome Size

697,578,683 base pairs arranged in 2n=20 chromosomes

Loci

34,496 loci containing protein-coding transcripts

Transcripts

36,338 protein-coding transcripts

FAQ

How was the genome sequenced?

Whole genome shotgun! What were you thinking?! Doesn't that produce a lousy genome sequence?

Although the first plant and animal genomes were sequenced by a BAC-by-BAC approach, almost all current animal and fungal genome sequencing projects use the whole genome shotgun strategy in which the entire genome is randomly sheared, subcloned, and redundantly sequenced. The ease, cost-efficiency, and speed of whole genome shotgun approach has made it the method of choice in many cases, but there are lingering concerns about its effectiveness for large repeat-rich plant genomes, especially grasses. Sorghum is the most complex plant genome sequenced to date by this strategy. Give it a spin. We hope you find it useful!

How was the assembly generated?

The Sbi1 release is a whole genome shotgun assembly produced by Jeremy Schmutz at JGI-Stanford Human Genome Center using the Arachne2 assembler in a mode tuned to the highly repetitive sorghum genome. The genome coverage is approximately 8x.

Why are there "super";s and chromosomes?

A "supercontig" (also known as a "super" or "scaffold") is a reconstructed genomic region that may contain modest gaps whose size is approximately known. Most of these supercontigs have been placed into 10 chromosome-size chunks, representing 90% of the genome and 99% of protein-coding regions.  

Is it complete?

Comparison with the physical and genetic map indicates that Sbi1 covers the vast majority of the available non-repetitive markers; comparison with the sorghum EST set suggests that more than 95% known sorghum protein-coding genes are represented in the assembly (many that aren't are turning out to be contamination of EST libraries). Both results support the claim that Sbi1 is largely complete with respect to "gene space."  You'll also find that vast tracts of repetitive sequence are also assembled. 

Is it accurate?

The vast majority of ESTs align to the genome at nearly 100% identity, suggesting that Sbi1 is highly accurate in genic regions. We are currently evaluating the base-pair-level accuracy in repetitive regions by comparing the assembly with BAC clones produced for the project. On a larger scale, we have identified approximately three dozen locations with apparent discrepancies between the shotgun assembly and the independently obtained maps. These will be reconciled in the chromosome-scale Sorbi1 release in spring 2007.

How do I find my favorite genes?

BLAST

To BLAST against the sorghum genome with peptide or nucleotide probes, here and select the Sorghum bicolor node on the tree. The default BLAST database is a sorghum genome assembly that has been masked for high fidelity repeats, and default BLAST parameters are suitable for use with grass peptides and coding sequences. You can view your blast alignment against the genome by clicking on the hit of interest to see the detailed alignment, and then clicking on the scaffold name (shown in blue). If you're interested in transposable element families in the sorghum genome, please DO NOT BLAST these, it'll just clog up our BLAST queue!  Similarly, please don't BLAST entire BACs.  

Search

We have pre-aligned known sorghum, maize, and sugarcane ESTs to the sorghum sequence, along with current proteomes of rice and Arabidopsis. If you enter into the Gbrowser "Search" box text keywords from common gene names like "zein" or "agamous", or gene identifiers like "At1g12340," the result will be a list of genomic regions that hit ESTs or rice/Arabidopsis genes that are associated with these words/identifiers. Clicking on the red diamonds will then bring you to the specific region of interest. Note that you may need to zoom in to see details, which are only shown over regions shorter than 50 kb.

Maize bins

by comparing (non-repetitive) sequence markers associated with "bins" on the maize genetic map, we have provisionally identified the syntenic sorghum regions for each maize bin. Note that since maize is paleotetraploid relative to sorghum, many genic regions of the sorghum genome are covered by two (and occasionally more) maize bins.

NOTE

The current "super" scaffolds bear no relation to the Sorbi0 release superscaffolds. To map forward, you can BLAST nucleotide or peptide sequences from the Sorbi0 release against the current genome.

How do I work with the sorghum genome browser?

How can I view the sorghum sequence and various genomic features?

The sorghum genome is available here. Detailed features are only visible when looking at 100 kb or smaller. You may need to zoom in to get to this size. Typically, clicking on a feature will reveal its sequence and alignment to the genome.

How do I retrieve sorghum sequence of interest to me?

From the browser, locate the region of interest. With your region in view, select "Download Sequence" from the menu above the Scroll/Zoom bar.  Then click the "Go" button and you'll get your sequence on your browser to cut and paste.  If you click on a gene model, you can retrieve the predicted peptide and coding sequencing.

What happens when I click on a gene on the browser

You'll see a web page that displays the predicted peptide, genomic span of the gene with coding exons shaded, and the (spliced) coding sequence.  From this page you can also launch BLAST vs. Phytozome organisms and gene families or the NCBI non-redundant protein database.

Where do the various tracks on the genome browser come from?

How were repeats identified?

The genome was masked using RepeatMasker. Nearly 66% of the genome appears to be covered by such clustered/over-represented regions. This is clearly an underestimate of the repeat content of sorghum, as many older/more diverged transposable element "fossils", as well as low copy elements, have not been characterized yet.

What is a SAMI?

"Sorghum assembled methyl-filtered islands" represent assemblies of methyl-filtered sorghum shotgun sequences, obtained from Pat Schnable's MAGI/SAMI analysis. These are enriched for genic regions but only cover portions of genes.

How were ESTs aligned?

We aligned the consensus EST sequences of sorghum, sugarcane, and maize from the TIGR Plantta database to the sorghum genome using Jim Kent's BLAT and NCBI BLAST.

How were rice and Arabidopsis peptides aligned?

The Arabidopsis and rice peptides were downloaded from NCBI RefSeq and aligned to the (unmasked) genome by gapped BLASTX; high-scoring sequence pairs (HSP's) are shown. Note that gapped BLAST was used to increase sensitivity, so that in many cases the HSP (shown in yellow) spans adjacent exons and the intervening intron(s). Also, small exons (evident from the maize/sorghum/sugarcane ESTs) are often missed.

How did you get the gene set for sorghum?

Where did the gene set come from?

Consensus gene predictions were built around several evidence sources. TIGR transcript assemblies were mapped on repeat-masked genome sequences, applying GenomeThreader with a splice site model of maize. Assemblies and ESTs of the following species were mapped: Allium cepa, Ananas comosus, Avena sativa, Brachypodium distachyon, Curcuma longa, Hordeum vulgare, Oryza sativa, Saccharum officinarum, Secale cereale, Sorghum bicolor, Sorghum halapense, Sorghum propinquum, Triticum aestivum, Zea mays, and Zingiber officinale. We also generated optimal spliced alignments (OSAs) as well as blastX alignments for a reference set of proteins consisting of the SWISSPROT database, Arabidopsis (TAIR6), Saccharomyces cerevisiae, and Rice (RAP2) proteomes. For each OSA, possible reading frames of size ³50 amino acids were collected as candidates for gene models. In addition, we identified gene models on repeat masked genomic sequences by ab initio methods (Fgenesh++, GeneID, GenomeScan/PASA). Next, we applied Jigsaw as a statistical combiner of all the supporting information above. A decision tree has been trained on a set of 987 gene models that have been edited by human supervision in the Apollo Genome Browser. All models, including those obtained from the first analysis series, were scored by blastp against the UniREF90 protein database and for each locus the best fitting model, i.e. the model with the highest bitscore, has been selected. In our final step, these predictions have been rerun through the PASA pipeline in order (i) to predict UTRs from maize, sorghum and sugarcane ESTs, (ii) to identify possible alternative splicing patterns and (iii) to fit all predicted models to the splice sites suggested by EST evidences of closely related species. This pipeline yielded 36,338 transcript models at 34,496 loci. In addition to the 28,003 complete models, we predicted 6493 candidate genes that lack a start and/or stop codon. These are therefore assigned as partial models. We only included such models in our annotation if they were not overlapping with complete predictions. Note that partial gene models may result from several, not mutually exclusive reasons: (i) sequencing or assembly errors may hinder both ab initio and homology based predictors to deduce a correct ORF; (ii) transposon activity may have truncated gene models; (iii) we have insufficient evidences from ab initio predictions or EST matches to provide a complete gene model.

How were UTRs identified in gene predictions?

The Program to assemble Spliced alignments, PASA (B. Haas), was run on the gene prediction set with all available sorghum ESTs. This produced 1842 alternatively spliced alignments and added UTR to 17,744 transcripts.

Why do models sometimes disagree with "obvious" exons from ESTs or homologous rice genes?

Two reasons. First, while annotation prediction programs does take homology information into account, they also adheres to an internal statistical model for what coding sequences in maize and related grasses "should" look like. So homology evidence may be “overriden” if it is inconsistent with expected codon usage, etc. A second and related problem is that ESTs are imperfect and sometimes grossly wrong, as they may include unspliced (retained) introns and/or genomic contamination of the cDNA library. By using a statistical model, gene predictors are able to reject such false data in some cases.

Why don't all the open reading frames (ORFs) start with methionine? Why don't all the ORFs end with a stop codon? How come my gene is only partially predicted?

GenomeScan is one of the better homology-based gene predictors available, but like all computational gene modeling algorithms, it is imperfect. Also, to avoid "run-on" models that inappropriately join adjacent genes, we only provided GenomeScan with our best guess for the genomic extent of a locus. If the statistical model of GenomeScan does not encounter what it believes to be the true start or end of a gene in our locus, the initial ATG or terminal stop codon may not be present in the model. So its partially GenomeScan's fault, and partially ours.