1 of 77

ga4gh.org

2 of 77

Sequence Annotation�Raiders of the Lost Schema

Evan Christensen

SA Working Group Coordinator

University of Utah

Sean Upchurch

SA Working Group Co-Chair

Caltech

Karen Eilbeck

SA Working Group Co-Chair

University of Utah

ga4gh.org

3 of 77

Sequence Annotation Is…

Sequence Annotation is developing a model for describing genomic features such as transcripts, gene, regulatory elements and related regions of interest

The model will enable sharing of knowledge associated with a location on the genome beyond the raw sequence at that location

ga4gh.org

4 of 77

Sequence Annotation Is Not…

The final answer to genome annotation

…but it does define a set of needed data

A data sharing API

…but it supports their design and implementation

A data format

…but it is an underlying data model

ga4gh.org

5 of 77

Sequence Annotation Model Scope

Provide a common framework for use by GA4GH and community.
Develop models for key concepts.
Allow models to be extended as needed by downstream development projects.

ga4gh.org

6 of 77

A comparison of data models relevant to sequence annotation

ga4gh.org

7 of 77

Data Models and Standards Development

Instantiated biological systems architecture and related data models, while initially meant to describe classes of biological entities, have in their own right become instances of biological knowledge differing from their conceptual counterparts across databases in meaningful ways.

ga4gh.org

8 of 77

Data Models and Standards Development

See what is already the same across major implementations. Then standardize what isn’t the same.

ga4gh.org

9 of 77

Data Models and Standards Development

Standardization should arise organically out of major implementations, taking the opportunity to address any areas for improvement but not starting from scratch.

ga4gh.org

10 of 77

What are sequence annotations?

Sequence annotations serve as a repository for prior knowledge, a tool to facilitate the exchange of new knowledge, and a foundation for analysis and interpretation.

ga4gh.org

11 of 77

GFF - The Generic (or General) Feature Format

seqid
source
type
start
end
score
strand
phase
attributes

https://github.com/The-Sequence-Ontology/Specifications/blob/master/gff3.md

ga4gh.org

12 of 77

What are sequence annotations?

Separate data models exist for sequence annotations at rest, in transit, and in use.

ga4gh.org

13 of 77

Sequence Annotations �at Rest

ga4gh.org

14 of 77

NCBI Data Model

ga4gh.org

15 of 77

NCBI Data Model

"We're using relational databases now because they're the current hot thing to use in computer science and there's lots of reasons for it but there's also a lot of limitations to that technology. There's no reason to believe that it'll be around forever, and so we don't want to lose everything we did in the relational databases, we want to provide ourselves a path where we've taken the information we've stored in a relational database and move it to the next generation database".  ��-Jim Ostell, NCBI, 1990

ga4gh.org

16 of 77

NCBI Data Model

GenBank flatfile

DNA-centered format
Features identified and associated with the sequence.
Sometimes related protein sequence appeared as annotated features or as a reference to another accession number but there was no direct connection between nucleic acid and protein in the data itself.

FEATURES Location/Qualifiers

source 1..813

/organism="Homo sapiens"

/mol_type="genomic DNA"

/isolate="65"

/db_xref="taxon:9606"

/sex="female"

/collected_by="Hala Alshwaikh"

gene <1..>813

/gene="BRCA1"

mRNA <1..>813

/gene="BRCA1"

/product="BRCA1"

CDS <1..>813

/gene="BRCA1"

/codon_start=3

/product="BRCA1"

/protein_id="WDY61431.1"

/translation="KPNEQTSKRHDSDTFPELKLTNAPGSFTKCSNTSELKEFVNPSL

PREEKEEKLETVXVSNNAEDPKDLMLSGERVLQTERSVESSSISLVPGTDYGTQESIS

LLEXSTLGKAKTEPNKCVSQCAAFENPKGLIHGCSKDNRNDTEGFKYPLGHEVNHSRE

TSIEMEESELDAQYLQNTFKVSKRQSFALFSNPGNAEEECATFSAHSGSLKKQSPKVT

FECEQKEENQGKNESNIKPVQTVNITAGFPVVGQKDKPVDNAKCSIKGGSRF"

exon <1..>813

/gene="BRCA1"

/number=10

ORIGIN

1 acaagccaaa tgaacagaca agtaaaagac atgacagyga tactttccca gagctgaagt

61 taacaaatgc acctggttct tttactaagt gttcaaatac cagtgaactt aaagaatttg

121 tcaatcctag ccttccaaga gaagaaaaag aagagaaact agaaacagtt raagtgtcta

181 ataatgctga agaccccaaa gatctcatgt taagtggaga aagggttttg caaactgaaa

241 gatctgtaga gagtagcagt atttcaytgg tacctggtac tgattatggc actcaggaaa

301 gtatctcgtt actggaagkt agcactctag ggaaggcaaa aacagaacca aataaatgtg

361 tgagtcagtg tgcagcattt gaaaacccca agggactaat tcatggttgt tccaaagata

421 atagaaatga cacagaaggc tttaagtatc cattgggaca tgaagttaac cacagtcggg

481 aaacaagcat agaaatggaa gaaagtgaac ttgatgctca gtatttgcag aatacattca

541 aggtttcaaa gcgccagtca tttgctctgt tttcaaatcc aggaaatgca gaagaggaat

601 gtgcaacatt ctctgcccac tctgggtcct taaagaaaca aagtccaaaa gtcacttttg

661 aatgtgaaca aaaggaagaa aatcaaggaa agaatgagtc taatatcaag cctgtacaga

721 cagttaatat cactgcaggc tttcctgtgg ttggtcagaa agataagcca gttgataatg

781 ccaaatgtag tatcaaagga ggctctaggt ttt

//

https://www.ncbi.nlm.nih.gov/nuccore/MW716260.1?report=genbank

ga4gh.org

17 of 77

NCBI Data Model

NCBI Data Model & ASN.1

Sought to design a specification without being tied to any particular computer platform or database technology.
Abstract Syntax Notation 1 (ASN.1; ISO 8824, 8825)
Each ASN.1 defined object has a matching "C" language structure.
Each "C" structure has at a minimum, a function to create it, write it to an ASN.1 stream, read it from an ASN.1 stream, and destroy it. Many objects have additional functions.

ga4gh.org

18 of 77

NCBI Data Model - Bioseq

ga4gh.org

19 of 77

NCBI Data Model - Bioseq

ga4gh.org

20 of 77

NCBI Data Model - Bioseq

ga4gh.org

21 of 77

NCBI Data Model - Bioseq

ga4gh.org

22 of 77

NCBI Data Model - Seq-inst

ga4gh.org

23 of 77

NCBI Data Model - Seq-inst

ga4gh.org

24 of 77

NCBI Data Model - Seq-inst

ga4gh.org

25 of 77

NCBI Data Model - Seq-loc

ga4gh.org

26 of 77

NCBI Data Model - Seq-annot

ga4gh.org

27 of 77

NCBI Data Model - Seq-annot

ga4gh.org

28 of 77

NCBI Data Model - Seq-annot

ga4gh.org

29 of 77

NCBI Data Model - Quick Summary

A container for biological sequence that:

Is instantiated as a particular type of sequence
Is uniquely identifiable
Can contain a descriptor containing database-specific metadata
Can contain sequence annotations (features, citations, etc.) and can associate those annotations with other bioseqs via Seq-locs and IDs.

ga4gh.org

30 of 77

Ensembl Core Data Model

ga4gh.org

31 of 77

Ensembl Core Data Model

Usually describes a program and some database that together are used to create a feature on a piece of sequence.
Each feature is marked with an analysis_id. The most important column is logic_name, which is used by the webteam to render a feature correctly on contigview (or even retrieve the right feature).
Logic_name is also used in the pipeline to identify the analysis which has to run in a given status of the pipeline.
The module column tells the pipeline which Perl module does the whole analysis, typically a RunnableDB module.

ga4gh.org

32 of 77

Ensembl Core Data Model

Stores information about sequence regions.
The primary key is used as a pointer into the dna table so that actual sequence can be obtained, and the coord_system_id allows sequence regions of multiple types to be stored.
Clones, contigs and chromosomes are all now stored in the seq_region table. Contigs are stored with the co-ordinate system 'contig'.
The relationship between contigs and clones is stored in the assembly table. The relationships between contigs and chromosomes, and between contigs and supercontigs, are stored in the assembly table.

ga4gh.org

33 of 77

Ensembl Core Data Model

dna

Contains DNA sequence.
This table has a 1:1 relationship with the seq_region table.

coord_system

Stores information about the available co-ordinate systems for the species identified through the species_id field.

ga4gh.org

34 of 77

Ensembl Core Data Model

The assembly table states, which parts of seq_regions are exactly equal.
It enables transformation of coordinates between seq_regions.
Typically this contains how chromosomes are made of contigs, clones out of contigs, and chromosomes out of supercontigs.
It allows you to artificially chunk chromosome sequence into smaller parts.
The data in this table defines the "static golden path", i.e. the best effort draft full genome sequence as determined by the UCSC or NCBI (depending which assembly you are using).
Each row represents a component, e.g. a contig, (comp_seq_region_id, FK from seq_region table) at least part of which is present in the golden path.

ga4gh.org

35 of 77

Ensembl Core Data Model

transcript

Stores information about transcripts.
Has seq_region_start, seq_region_end and seq_region_strand for faster retrieval and to allow storage independently of genes and exons.

ga4gh.org

36 of 77

Ensembl Core Data Model

gene

Allows transcripts to be related to genes.

gene_attrib

Enables storage of attributes that relate to genes.

ga4gh.org

37 of 77

Ensembl Core Data Model

protein_feature

Stores translation alignments generated from Blast (or Blast-like) comparisons.

ga4gh.org

38 of 77

Ensembl Core Data Model

translation

Describes which parts of which exons are used in translation.
Transcripts are related to translations by the transcript_id key in this table.

ga4gh.org

39 of 77

Ensembl Core Data Model

exon

Stores data about exons.
Associated with transcripts via exon_transcript.
Allows access to contigs seq_regions.

ga4gh.org

40 of 77

Ensembl Core Data Model

biotype

Stores data about biotypes and mappings to Sequence Ontology.

xref

Holds data about objects which are external to EnsEMBL, but need to be associated with EnsEMBL objects.
Information about the database that the external object is stored in is held in the external_db table entry referred to by the external_db column.

ga4gh.org

41 of 77

Ensembl Core Data Model

external_db

Stores data about the external databases in which the objects described in the xref table are stored.

ga4gh.org

42 of 77

Ensembl Core Data Model

Stores data about the external databases in which the objects described in the xref table are stored.

ga4gh.org

43 of 77

Ensembl Core Data Model - Quick Summary

Primarily focused on sequence annotations (features) that can refer back to sequence via the relational schema.
Relevant metadata is contained in specialized tables.
Sequence features have a location, are identified using sequence ontology terms, and are associated with specific analyses.

ga4gh.org

44 of 77

Alliance of Genome Resources

ga4gh.org

45 of 77

Alliance of Genome Resources

Several model organism databases merging into one unified schema.
Will eventually be managed with PostgreSQL.
Are still working on modeling sequence features.
We’re starting to look at FlyBase (GMOD) and WormBase (AceDB).

ga4gh.org

46 of 77

Generic Model Organism Database (GMOD)

How is a “central dogma” gene represented in Chado?

ga4gh.org

47 of 77

Alliance of Genome Resources - Brief Summary

Standards developed at GA4GH could potentially be relevant to standardization efforts at AGR.
Chado (FlyBase)

Centers on sequence features (Sequence Ontology and Ensembl).
Considers genes and transcripts as sequence features.

AceDB (WormBase)

Claims to handle sequence “better”
Potential example for establishing model extension guidelines (tight feedback loop between curators and subject matter experts).

ga4gh.org

48 of 77

Sequence Annotations �in Transit

ga4gh.org

49 of 77

INSDC Feature Table

Feature table format example (EMBL):

FT source 1..1859

FT /db_xref="taxon:3899"

FT /organism="Trifolium repens"

FT /tissue_type="leaves"

FT /clone_lib="lambda gt10"

FT /clone="TRE361"

FT /mol_type="genomic DNA"

FT CDS 14..1495

FT /db_xref="MENDEL:11000"

FT /db_xref="UniProtKB/Swiss-Prot:P26204"

FT /note="non-cyanogenic"

FT /EC_number="3.2.1.21"

FT /product="beta-glucosidase"

FT /protein_id="CAA40058.1"

FT /translation="MDFIVAIFALFVISSFTITSTNAVEASTLLDIGNLSR.......

---------+---------+---------+---------+---------+---------+---------+---------

1 10 20 30 40 50 60 70 79

Feature table format example (GenBank):

source 1..8959

/organism="Homo sapiens"

/db_xref="taxon:9606"

/mol_type="genomic DNA"

gene 212..8668

/gene="NF1"

CDS 212..8668

/gene="NF1"

/note="putative"

/codon_start=1

/product="GAP-related protein"

/protein_id="AAA59924.1"

/translation="MAAHRPVEWVQAVVSRFDEQLPIKTGQQNTHTKVSTE.......

---------+---------+---------+---------+---------+---------+---------+---------

1 10 20 30 40 50 60 70 79

Feature table format example (DDBJ):

source 1..2136

/clone="pK28"

/organism="Rattus norvegicus"

/strain="Sprague-Dawley"

/tissue_type="kidney"

/mol_type="genomic DNA"

mRNA 19..2128

CDS 31..1212

/codon_start=1

/function="Dual specificity protein tyrosine/threonine

kinase"

/product="MAP kinase kinase"

/protein_id="BAA02603.1"

/translation="MPKKKPTPIQLNPAPDGSAVNGTSSAETNLEALQKKL.......

---------+---------+---------+---------+---------+---------+---------+---------

1 10 20 30 40 50 60 70 79

https://www.insdc.org/submitting-standards/feature-table/

ga4gh.org

50 of 77

Sequence Ontology

ga4gh.org

51 of 77

Sequence Annotations �in Use

ga4gh.org

52 of 77

GFF - The Generic (or General) Feature Format

seqid
source
type
start
end
score
strand
phase
attributes

https://github.com/The-Sequence-Ontology/Specifications/blob/master/gff3.md

ga4gh.org

53 of 77

SA Model v0.1a

ga4gh.org

54 of 77

SA Model v0.1a

ga4gh.org

55 of 77

SA Model v0.1a

The relation/context entity (actual name TBD) will enable users to associate multiple sequences with one feature, multiple features with one sequence, etc. This will be key to our representation of alternatively spliced transcripts and gene models.

ga4gh.org

56 of 77

Levels of Interoperability

System interoperability: �Operating system and hardware required to run the system.
Structural interoperability: �Database Architecture - Defines how the system stores and utilizes its data for operations.
Syntactic interoperability: �Formats - Represents descriptions of data representation in a machine-readable format.
Semantic interoperability: �Content - Enables machine-driven parsing and interpretation of the data's intended meaning.

Costin A, Eastman C. Need for Interoperability to Enable Seamless Information Exchanges in Smart and Sustainable Urban Systems. J. Comput. Civ. Eng 2019 May;33(3):04019008.

ga4gh.org

57 of 77

Levels of Interoperability

The INSDC essentially achieved system, structural, and syntactic interoperability in the early 90s.�

True semantic interoperability still eludes us though.

Costin A, Eastman C. Need for Interoperability to Enable Seamless Information Exchanges in Smart and Sustainable Urban Systems. J. Comput. Civ. Eng 2019 May;33(3):04019008.

ga4gh.org

58 of 77

Semantic Interoperability

How well does your model handle new knowledge not considered in your initial design?

Data models need a “Discovery” component.
Extensions to data models may not be keeping up with user needs.
As data models become standardized and more stakeholders demand more of shared schemas, how will models extend to accommodate their domain while maintaining the shared conceptual overlap such initiatives aim to achieve?
Researchers need to be able to transmit novel findings immediately, potentially using natural language or free text.

ga4gh.org

59 of 77

Potential Follow-up Study

Quantify, measure, and characterize convergence amongst the extensions to annotation providers’ data models representing sequence annotations at rest, in transit, and in use. �

We hypothesize that there are commonalities in the ways annotation providers have had to extend their models over time as the body of scientific work has grown.
If so, this would lend strong support for a prioritizing certain ways of standardizing our representation of genomic knowledge within and between information systems.

ga4gh.org

60 of 77

Potential Follow-up Study

Compare the internal validity of data and metadata served by the major annotation providers against the data definitions and logical constructs specified by their respective data models. �

Some data may no longer be alignment with or valid within their own constructs simply due to the long-term usage of these data models despite incremental extensions over time.
Patterns in inconsistencies between data, metadata, and data models, identified with advanced computational methods, will inform recommendations for an improved, standardized approach to representing genomic knowledge at scale.

ga4gh.org

61 of 77

Discussion

Come to an agreement on the core SA model objects and their relation to other GKS objects (e.g. VRS Location, VA Evidence Lines).
Reestablish scope for next steps.
Discuss potential driver projects that could lead towards a project proposal for a v1.0 GA4GH SA data model standard.
Identify items for further research and development that would segue into those potential driver projects.

ga4gh.org

62 of 77

Appendix

ga4gh.org

63 of 77

Data Models (What and Why?)

COMMUNICATION

Can represent various levels of abstraction. Facilitates a deeper understanding of what’s being designed.

ORGANIZATION

Organizes data elements. Built to address specific needs.

STANDARDIZATION

Standardizes relationships between data elements and real-world entities.

ga4gh.org

64 of 77

Data Models (What and Why?)

Physical

Provide a finalized schema (design) that can be implemented.

3

Logical

Provide greater detail about the concepts and relationships under consideration. Follows a specific notation style.

2

Conceptual

�Offer a 10,000 ft. view of the contents, organization, and rules involved with the system.

1

ga4gh.org

65 of 77

NCBI Data Model

Nuc-prot set: A collection of a DNA sequence and its translation products.
Seg-set: contains segmented sequences, all of their connections, and features. There may be gaps and/or overlaps between segments (contigs).
NCBI Software Toolkit has functions that gather data from a segmented sequence “on the fly” and can provide graphical views, GenBank flatfile views, FASTA views, etc.
4 Core Data Elements: Bibliographic citations, DNA sequences, protein sequences, 3D structures.