Assembling Genes from Predicted Exons in Linear Time with Dynamic Programming

Abstract

In a number of programs for gene structure prediction in higher eukaryotic genomic sequences, exon

Dandruff-associated Malassezia genomes reveal convergent and divergent virulence traits shared with plant and human fungal pathogens

“…The genome was assembled by using Phrap. Gene models were constructed by combining sequencing information from 1,536 EST sequences and in silico predictions from TWIN-SCAN (31), GeneID (32), and GeneWise (33) (SI Text). Putative gene function assignments were obtained through homology searches against the National Center for Biotechnology Information nr, PFAM (34), and TIGRFAM (35) databases.…”

Sequencing of allotetraploid cotton (Gossypium hirsutum L. acc. TM-1) provides a resource for fiber improvement

“…GeneWise 56 was used to predict the exact gene structure of the corresponding genomic regions on each BLAST hit. Five ab initio gene prediction programs, Augustus (version 2.5.5) 57 , Genscan (version 1.0) 58 , GlimmerHMM (version 3.0.1) 59 , Geneid 60 , and SNAP 61 , were used to predict coding regions in the repeat-masked genome. Finally, RNA-seq data were mapped to the assembly using Tophat (version npg PEG, heat and cold.…”

GeneID in Drosophila

“…We tested two additional models of coding DNA before deciding for a Markov model of order 5, a Codon usage model, and a model that combined a Markov model of order 1 of the translated amino acid sequence and a Codon preference model (see Guigó 1999 for details on these models). In both cases, log-likelihood ratios were obtained in a similar way to the Markov model loglikelihood ratios (see Methods).…”

“…This new version maintains the hierarchical structure (signal to exon to gene) in the original GeneID, but we have simplified the scoring schema and furnished it with a probabilistic meaning: Scores for both exon-defining signals and protein-coding potential are computed as loglikelihood ratios, which for a given predicted exon are summed up into the exon score, in consequence also a log-likelihood ratio. Then, a dynamic programming algorithm (Guigó 1998) is used to search the space of predicted exons to assemble the gene structure (in the general case, multiple genes in both strands) maximizing the sum of the scores of the assembled exons, which can also be assumed to be a log-likelihood ratio. Execution time in this new version of GeneID grows linearly with the size of the input sequence, currently at ∼2 Mb per minute in a Pentium III (500 MHz) running linux.…”