GENSCAN

GENSCAN
Name	GENSCAN
Author	Matthew Burge; Steven Karlin
Developer	Burge & Karlin Laboratory; Washington University in St. Louis
Released	1997
Latest release	legacy
Operating system	Cross-platform
Genre	Bioinformatics; Computational biology
License	Academic

GENSCAN GENSCAN is a computational gene prediction program developed for identifying protein-coding genes in genomic DNA sequences. It was introduced to the bioinformatics community to predict exon–intron structures in eukaryotic genomes using probabilistic models and has been widely cited in genomic projects, comparative genomics, and annotation pipelines.

Overview

GENSCAN was created within the context of large-scale sequencing efforts such as the Human Genome Project and used in projects involving organisms including Homo sapiens, Mus musculus, Drosophila melanogaster, and various plant genomes. The tool addressed needs that also motivated software like Genscan II? and contemporaries such as Augustus (gene prediction), Grail, FGENESH, GeneMark, and EuGene during a period when databases such as GenBank, UniProt, and resources like Ensembl and UCSC Genome Browser were expanding. Development and evaluation interacted with consortia and initiatives including the National Center for Biotechnology Information, Wellcome Trust Sanger Institute, Joint Genome Institute, and annotation efforts by projects analogous to the ENCODE Project.

Algorithm and Models

GENSCAN’s core relies on hidden Markov models (HMMs) and generalized hidden Markov models (GHMMs) related to approaches used in tools like HMMER and methods developed in computational linguistics and signal processing. The implementation models canonical features such as 5' splice sites and 3' splice sites, promoter-like signals, start codons (ATG), stop codons (TAA, TAG, TGA), and codon usage patterns resembling analyses from studies published in venues like Nature, Science, and Genome Research. Training strategies were influenced by statistical methods discussed in work associated with researchers at Stanford University, Massachusetts Institute of Technology, University of California, Berkeley, and University of Washington. GENSCAN integrates parameter estimation comparable to methods seen in Expectation–Maximization algorithm applications used by groups at EMBL-EBI and laboratories collaborating via European Molecular Biology Laboratory networks.

Input, Output, and Usage

GENSCAN accepts raw DNA sequences typically in FASTA format, similar to inputs used by tools such as BLAST, Clustal, MAFFT, and Bowtie. Users provide sequence with organism model selection comparable to specifying models in GeneMark.hmm or choosing parameter sets as in Augustus. Output includes predicted exon and intron coordinates, predicted protein translations, and scores analogous to outputs from GlimmerHMM and annotation tools integrated into pipelines like those used by RefSeq curation. The software was distributed as command-line binaries and integrated into workflow managers similar to Galaxy (platform), Cufflinks pipelines, and pipeline frameworks used at institutions such as Broad Institute and Cold Spring Harbor Laboratory.

Performance and Accuracy

Benchmarks compared GENSCAN predictions against curated gene sets from projects carried out at Sanger Institute, NCBI RefSeq, and organism-specific databases like FlyBase, TAIR, and MGI. Performance metrics such as sensitivity, specificity, exon-level accuracy, and nucleotide-level accuracy were assessed in studies published in journals including Nucleic Acids Research, Bioinformatics, and Genome Biology. Comparisons often involved tools such as FGENESH, AUGUSTUS, GeneMark-ES, and SNAP, and evaluations used assemblies from sequencing platforms like Illumina, Roche 454, and earlier capillary sequencers from Applied Biosystems. GENSCAN exhibited strengths in predicting typical gene structures in vertebrates but showed lower accuracy in predicting alternatively spliced isoforms relative to methods incorporating RNA-seq evidence produced by technologies like RNA-Seq and analytical tools such as TopHat and STAR.

Applications and Impact

GENSCAN contributed to early annotations in projects tied to Human Genome Project publications, comparative studies involving chimpanzee and mouse genomes, and gene discovery in model organisms such as Arabidopsis thaliana and Caenorhabditis elegans. It influenced downstream resources and databases including GenBank, RefSeq, UniProtKB/Swiss-Prot, InterPro, and annotation tracks in the UCSC Genome Browser. Educational and research uses appeared in classrooms at universities including Harvard University, University of Cambridge, and Princeton University as an exemplar of probabilistic gene prediction. Its conceptual framework informed later software development in academic labs at Johns Hopkins University, Yale University, Cold Spring Harbor Laboratory, and corporate research at companies like Illumina, Inc. and Thermo Fisher Scientific.

Limitations and Development History

GENSCAN’s models were trained on species-specific parameter sets and struggled with genomes having unusual GC content, pervasive alternative splicing like in Drosophila melanogaster, or compact prokaryote-like gene structures explored in projects at JGI and studies comparing eukaryotic and prokaryotic gene architectures. As transcriptomics technologies matured, methods leveraging expressed sequence tags from projects like ESTs and RNA-seq became preferred; pipelines incorporating tools such as Cufflinks, StringTie, MAKER (genome annotation), and evidence combiners like EVM (EVidenceModeler) superseded single-model predictors. Subsequent research and software—refinements exemplified by AUGUSTUS improvements, BRAKER automation, and community-driven annotation efforts at institutions like Ensembl—built upon the conceptual foundations laid by GENSCAN while addressing its limitations.

Category:Bioinformatics software