Sequence Annotation Pipeline

Annotation vagaries and inconsistencies are a major impediment to sequence analysis. Virus Variation mitigates this problem using standardized sequence annotation pipelines that provide consistent annotation across all sequences belonging to a given viral species. Reference sequence sets are used to annotate proteins and other biologically and clinically relevant features. For example, the flu annotation pipeline generates information about drug-resistance mutations and the completeness of nucleotide and coding region sequences; both are stored in the database.

In general, the pipelines for each virus loaded into Virus Variation use a common backbone but unique reference protein sets and parsing strategies. For example, in the Dengue annotation pipeline the incoming sequence is initially assigned a genotype using megaBlast and a reference sequence set. That genotype assignment then points the annotation pipeline to a specific set of reference proteins that are used to annotate the new sequence. The annotation pipelines are used for both internal database loading and as a public resource for Influenza virus—providing standardized annotation for some GenBank submissions.

GenBank Submission Pipeline for Influenza Viruses

NCBI is a collaborator with the NIAID Influenza Genome Sequencing Project and has been tasked with gathering Influenza sequences and related metadata from J. Craig Venter Institute (JCVI), annotating the sequences, and releasing them in GenBank. NCBI has created an automated pipeline to facilitate the large number of sequences generated from the project.

In the pipeline, metadata are retrieved and updated daily from a JCVI ftp site and loaded onto an internal NCBI database. NCBI works closely with JCVI, viral sample providers, and the influenza virus research community in establishing the minimum and optional metadata sets to be incorporated into GenBank records. Organism names for new virus isolates are entered in the NCBI Taxonomy Database. NCBI staff also manually review the metadata and communicate with data providers if there are any issues.

Sequencing data are assembled at JCVI and consensus sequences are verified with the Influenza Virus Genome Annotation Tool (FLAN, for FLu Annotation, see the Virus Genome Processing and Tools Chapter) and then submitted to NCBI through ftp once they are error free. At NCBI, the sequences are processed by FLAN and feature tables generated. These are combined with associated metadata to create GenBank files. In the past 8 years, nearly 11,200 complete influenza virus genomes have been generated from the NIAID project, and published in GenBank.

Database Loading Pipeline

The database loading pipeline is an automated process that parses data from records available in GenBank and maps them to fields used in the Virus Variation database. This process uses generalized parsing strategies to capture both common biological data like host and country of origin, as well as individualized strategies to capture more specific—often clinically relevant—data associated with particular viruses.

The loading pipelines are dependent on vocabulary lists that allow mapping of data parsed from records to controlled descriptors used within the database and displays. For example, host names—including common names and misspelled names—are mapped with these vocabulary lists to scientific names associated with taxonomy IDs in the NCBI Taxonomy Database and host group names like “birds” or “mammals” used in the search pages.

These automated processes are augmented by manual operations based on literature and semi-automated procedures used to capture third-party data releases. Annotation and data capture is also facilitated by community outreach efforts that seek to develop standard, experimentally-driven gene models and reference protein sets. These efforts also encourage the inclusion of rich metadata sets in public database submissions, as well as metadata sharing.

Database

The Virus Variation database stores sequence information derived from the annotation pipeline and associated metadata describing the sample in standardized formats. To balance storage flexibility with efficient data retrieval, Virus Variation combines a relational database with documents containing raw data.

Curation Interface

Since the Virus Variation database loading procedure uses a hybrid of automated and manual procedures, it is important that NCBI staff have the ability to review sequences with loading errors as well as enter data manually into the database. The Virus Variation curation interface enables curators to filter and sort sequences based on virus type, loading errors, and on a number of descriptors like sequence length. A number of editable fields are displayed for each sequence—such as country, isolation date, and host—in a generalized format that is the same for each virus. Curators can review data associated with a given sequence and enter data manually into these fields as guided by literature or other sources. Additionally, there is the ability to adjust the displayed fields and messages to fit the needs of specific viruses and/or database loading procedures.

Search Interface

The unique search interface allows users to construct database queries based on a number of criteria including gene or protein region, GenBank accessions, and keywords, as well as biologically relevant descriptors like disease associations, host organism, and geographic information about the sample. Although the same basic interface design is used throughout the resource, the interface is customized to include specific search fields for individual viruses. The number of sequence records retrieved by a search is displayed with the query builder frame of the page—so that the user can modify search parameters. Once the desired query is built, retrieved sequences can be downloaded in a variety of formats directly or can be displayed within the results page.

Figure 1.

The Virus Variation search interface. Users can use a number of search criteria including sequence patterns, host, geographic region, and collection date to retrieve either protein or DNA sequences from specified genome regions.

Results Page

The results page displays all the sequences retrieved in a given search where individual sequences can be selected prior to subsequent analysis or download. Individual records can be sorted by a variety of descriptors, selected or deselected, downloaded, sent to the multi-sequence alignment viewer, or sent to the phylogenetic tree viewer.

Figure 2.

The Virus Variation search results page. Records retrieved during a search can be displayed within the results page where individual sequences can be selected for download or further analysis.

Multi-sequence Alignment Viewer

The multi-sequence alignment viewer allows users to display alignments of selected protein or nucleotide sequences. Alignments are precalculated to save processing time. The viewer is based on the Genome Workbench alignment viewer and includes a number of advanced features including multiple display and scoring options. In the default view a consensus sequence is displayed as the anchor, and a reference feature table is used to define protein (or gene) positions and other important landmarks. The displayed features facilitate navigation along the alignment and allow users to hone in on regions of interest. The anchor sequence can be changed from a consensus sequence to any in the alignment to facilitate greater scrutiny of specific sequences within the alignment.

Figure 3.

The Virus Variation multiple sequence alignment viewer. Selected protein or nucleotide sequences can be displayed in precalculated alignments allowing rapid comparison of sequences.

Phylogenetic Tree Viewer

The phylogenetic tree viewer displays phylogenetic trees built from alignments of sequences selected in the results page. The current viewer includes collapsible leaves, which allows a user to adjust the resolution of a selected subtree to improve viewing of large data sets (3). Users can also markup sequences based on date range and search for and tag sequences based on country, host, accession, and other descriptors. This feature provides a graphic representation of the metadata associated with sequences, enhancing the user’s ability to make associations between phylogenetics and sequence descriptors.

Figure 4.

The Virus Variation phylogenetic tree viewer. Selected nucleotide and protein sequences can be quickly displayed on phylogentic trees using precalculated alignments and a variety of clustering and distance algorithms. Sequences can be searched by metadata, such as country, host, and accession, and marked up in green. Sequences from specific dates can also be highlighted in red.