From 0351f51167e134fbdfb5dce295d6bbf2d0c33fb9 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 13 Nov 2024 10:29:17 -0500 Subject: [PATCH 1/9] fix github links --- docs/decision_record.md | 65 +++++++++++++++++++++-------------------- 1 file changed, 34 insertions(+), 31 deletions(-) diff --git a/docs/decision_record.md b/docs/decision_record.md index 7e57d31..fe1fdcb 100644 --- a/docs/decision_record.md +++ b/docs/decision_record.md @@ -8,6 +8,9 @@ The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "S [TOC] +## 2024-11-13 Attributes can be designed as `passthru` or `transient`. + + ## 2024-10-02 The `/collection` and `/attribute` endpoints will both be `REQUIRED` @@ -96,7 +99,7 @@ In the future if the number of proposed ancillary attributes grows, it could mov ### Linked issues -- +- ## 2024-02-21 We will specify core sequence collection attributes and a process for adding new ones @@ -120,9 +123,9 @@ Choosing to host this list as a list of issues allows the list to always be up t ### Linked issues - - - - - - + - + - + - ## 2024-01-10 Clarifications on the purpose and form of the JSON schema in service-info @@ -148,8 +151,8 @@ Another issue is that we wanted the schema to be a place where a user could see ### Linked issues -- -- +- +- ## 2024-01-06 The comparison function use more descriptive attribute names @@ -171,7 +174,7 @@ The comparison function is designed to compare two sequence collections by inter ### Linked issues -- +- ## 2023-08-25 The user-facing API will neither expect nor provide prefixes @@ -236,7 +239,7 @@ properties: ### Linked issues -- https://github.com/ga4gh/seqcol-spec/issues/40 +- https://github.com/ga4gh/refget/issues/40 ## 2023-07-26 There will be no metadata endpoint @@ -256,9 +259,9 @@ We distinguished between two types of metadata: ### Linked issues -- -- -- +- +- +- ## 2023-07-12 - Required attributes are: lengths and names @@ -302,7 +305,7 @@ This leads us to the conclusion that *sequences* should be optional, and *names* ### Linked issues -- +- ## 2023-06-14 - Internal digests SHOULD NOT be prefixed @@ -335,7 +338,7 @@ Adding prefixes will complicate things and does not add benefits. Prefixes may b ### Linked issues -- +- ## 2023-06-28 - SeqCol JSON schema defines reserved attributes without additional namespacing @@ -400,7 +403,7 @@ Thus, we introduce the idea of *inherent* vs *non-inherent attributes*. Inherent ### Linked issues -- +- ### Alternatives considered @@ -420,7 +423,7 @@ While non-ASCII array names would be compatible with our current specification, ### Linked issues -- +- ## 2023-01-25 - The digest algorithm will be the GA4GH digest @@ -449,7 +452,7 @@ Under this scheme the string `ACGT` will result in the `sha512t24u` digest `aKF4 ### Linked issues -- [https://github.com/ga4gh/seqcol-spec/issues/30](https://github.com/ga4gh/seqcol-spec/issues/30) +- [https://github.com/ga4gh/refget/issues/30](https://github.com/ga4gh/refget/issues/30) ## 2023-01-12 - How sequence collection are serialized prior to digestion @@ -536,9 +539,9 @@ It also future-proofs the serialisation method if we ever allow complex object t ### Linked issues - - [https://github.com/ga4gh/seqcol-spec/issues/1](https://github.com/ga4gh/seqcol-spec/issues/1) - - [https://github.com/ga4gh/seqcol-spec/issues/25](https://github.com/ga4gh/seqcol-spec/issues/25) - - [https://github.com/ga4gh/seqcol-spec/issues/33](https://github.com/ga4gh/seqcol-spec/issues/33) + - [https://github.com/ga4gh/refget/issues/1](https://github.com/ga4gh/refget/issues/1) + - [https://github.com/ga4gh/refget/issues/25](https://github.com/ga4gh/refget/issues/25) + - [https://github.com/ga4gh/refget/issues/33](https://github.com/ga4gh/refget/issues/33) ### Known limitations @@ -636,7 +639,7 @@ We should be consistent by using these terms to refer to the above representatio ### Linked issues -- +- ## 2022-06-15 - Structure for the return value of the comparison API endpoint @@ -704,8 +707,8 @@ The primary purpose of the compare function is to provide a high-level view of h ### Linked issues -- -- +- +- ### Alternatives considered @@ -778,8 +781,8 @@ We need a formal definition of a sequence collection. The schema provides a mach ### Linked issues -- -- +- +- ## 2021-12-01 - Endpoint names and structure @@ -825,8 +828,8 @@ For the `POST comparison` endpoint, we made 2 limitations to simplify the implem ### Linked issues -- [https://github.com/ga4gh/seqcol-spec/issues/21](https://github.com/ga4gh/seqcol-spec/issues/21) -- [https://github.com/ga4gh/seqcol-spec/issues/23](https://github.com/ga4gh/seqcol-spec/issues/23) +- [https://github.com/ga4gh/refget/issues/21](https://github.com/ga4gh/refget/issues/21) +- [https://github.com/ga4gh/refget/issues/23](https://github.com/ga4gh/refget/issues/23) ## 2021-09-21 - Order will be recognized by digesting arrays in the given order, and unordered digests will be handled as extensions through additional attributes @@ -854,7 +857,7 @@ To conclude, option A seems simple and straightforward, satisfies for a basic im ### Linked issues -- https://github.com/ga4gh/seqcol-spec/issues/5 +- https://github.com/ga4gh/refget/issues/5 ### Known limitations @@ -877,7 +880,7 @@ However, there are also scenarios for which the order of sequences in a collecti ### Linked issues -- [https://github.com/ga4gh/seqcol-spec/issues/5](https://github.com/ga4gh/seqcol-spec/issues/5) +- [https://github.com/ga4gh/refget/issues/5](https://github.com/ga4gh/refget/issues/5) ### Known limitations @@ -917,8 +920,8 @@ This will allow retrieving individual attributes, and testing for identity of in ### Linked issues -- [https://github.com/ga4gh/seqcol-spec/issues/8#issuecomment-773489450](https://github.com/ga4gh/seqcol-spec/issues/8#issuecomment-773489450) -- [https://github.com/ga4gh/seqcol-spec/issues/10](https://github.com/ga4gh/seqcol-spec/issues/10) +- [https://github.com/ga4gh/refget/issues/8#issuecomment-773489450](https://github.com/ga4gh/refget/issues/8#issuecomment-773489450) +- [https://github.com/ga4gh/refget/issues/10](https://github.com/ga4gh/refget/issues/10) ### Known limitations @@ -937,7 +940,7 @@ Should a wider GA4GH standard appear from [TASC issue 5](https://github.com/ga4g ### Linked issues -- [https://github.com/ga4gh/seqcol-spec/issues/2](https://github.com/ga4gh/seqcol-spec/issues/2) +- [https://github.com/ga4gh/refget/issues/2](https://github.com/ga4gh/refget/issues/2) ### Known limitations From b1dc384a56ec7fc01c71d4abed03f155619e4bd2 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 13 Nov 2024 12:13:29 -0500 Subject: [PATCH 2/9] write up recent decisions and changes --- docs/decision_record.md | 36 +++++++ docs/seqcol.md | 230 +++++++++++++++++++++++++++------------- 2 files changed, 195 insertions(+), 71 deletions(-) diff --git a/docs/decision_record.md b/docs/decision_record.md index fe1fdcb..1cbb184 100644 --- a/docs/decision_record.md +++ b/docs/decision_record.md @@ -10,7 +10,43 @@ The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "S ## 2024-11-13 Attributes can be designed as `passthru` or `transient`. +### Decision + +We add two new attribute qualifiers: transient and passthru. + +- Passthru attributes are not digested in transition from level 2 to level 1. Most attributes of the canonical (level 2) seqcol representation are digested to create the level 1 representation. But sometimes, we have an attribute for which digesting makes little sense. These attributes are passed through the transformation, so they show up on the level 1 representation in the same form as the level 2 representation. Thus, we refer to them as passthru attributes. +Transient attributes + +- Transient attributes are not retrievable from the attribute endpoint. Most attributes of the sequence collection can be retrieved through the /attribute endpoint. However, some attributes may not be retrievable. For example, this could happen for an attribute that we intend to be used primarily as an identifier. In this case, we don't necessarily want to store the original content that went into the digest into the database, because it might be redundant. We really just want the final attribute. These attributes are called transient because the content of the attribute is no longer stored and is therefore no longer retrievable. + +### Rationale + +As we worked on more advanced attributes, and with the addition of the `/attribute` endpoint, we realized these changes necessitate a bit more power for the schema to specify behavior of the attributes. For the basic seqcol attributes (names, lengths, sequences) and original endpoint, the general algorithm and basic qualifiers (required, inherent, collated) suffice to describe the representation. But some more nuanced attributes require additional qualifiers to describe their intention and how the server should be behave for the `/attribute` endpoint. For example, sorted_name_length_pairs and sorted_sequences are intended to provide alternative tailored identifiers and comparisons, and not necessarily useful for independent attribute lookup. Similarly, custom extra attributes, like author or alias, may be simple appendages that don't need the complex digesting procedure we use for the basic attributes. In order to flag such attributes in a way that can govern slightly different server expectations, we need a couple of additional advanced attribute qualifiers. For this purpose, we added the passthru and transient qualifiers. + +### Linked issues + +- + + +## 2024-10-02 Minimal schema should now require sequences, and lengths should not be inherent. + +### Decision + +We will update the minimal schema with these changes: 1. Move sequences into 'required', and 2. remove lengths from 'inherent'. So the final qualifiers would be: +- required: names, lengths, and sequences +- inherent: names, sequences + + +### Rationale + +Originally, there was a good rationale for making sequences not required, to allow for coordinate systems to be represented as a seqcol. +But with the new `/attribute` endpoint, there's a better way to handle it, using `name_length_pairs` and `sorted_name_length_pairs` attributes. +Then, with sequences required, it does not make sense for lengths to be inherent because they are computable from sequences. +So essentially, the attribute endpoint allows us to move away from handling coordinate systems as top-level entities, and instead moves toward using the attribute endpoint for coordinate systems. + +### Linked issues +- ## 2024-10-02 The `/collection` and `/attribute` endpoints will both be `REQUIRED` diff --git a/docs/seqcol.md b/docs/seqcol.md index 5daa2b9..64d6eeb 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -30,7 +30,7 @@ A common example and primary use case of sequence collections is for reference g In brief, the project specifies several procedures: 1. **An algorithm for encoding sequence collection identifiers.** The GA4GH standard [refget sequences](http://samtools.github.io/hts-specs/refget.html) specifies a way to compute deterministic sequence identifiers from individual sequences. Seqcol uses refget sequence identifiers and adds functionality to wrap them into collections of sequences. Seqcol also handles sequence attributes, such as their names, lengths, or topologies. Seqcol digests are defined by a hash algorithm, rather than an accession authority, and are thus decentralized and usable for private sequence collections, cases without connection to a central database, or validation of sequence collection content and provenance. -2. **An API describing lookup and comparison of sequence collections.** Seqcol specifies a RESTful API to retrieve the sequence collection given a digest. A main use case is to reproduce the exact sequence collection (*e.g.* reference genome) used for analysis, instead of guessing based on a human-readable identifier. Seqcol also provides a standardized method of comparing the contents of two sequence collections. This comparison function can *e.g.* be used to determine if analysis results based on different references genomes are compatible. +2. **An API describing lookup and comparison of sequence collections.** Seqcol specifies an http API to retrieve the sequence collection given a digest. A main use case is to reproduce the exact sequence collection (*e.g.* reference genome) used for analysis, instead of guessing based on a human-readable identifier. Seqcol also provides a standardized method of comparing the contents of two sequence collections. This comparison function can *e.g.* be used to determine if analysis results based on different references genomes are compatible. 3. **Recommended ancillary, non-inherent attributes.** Finally, the protocol defines several recommended procedures that will improve the compatibility across Seqcol servers, and beyond. ## Use cases @@ -64,13 +64,14 @@ Some other examples of common use cases where the use of seqcol is beneficial in ## Definitions of key terms +### General terms + - **Alias**: A human-readable identifier used to refer to a sequence collection. - **Array**: An ordered list of elements. -- **Collated**: A qualifier applied to a seqcol attribute indicating that the values of the attribute matches 1-to-1 with the sequences in the collection and are represented in the same order. - **Coordinate system**: An ordered list of named sequence lengths, but without actual sequences. - **Digest**: A string resulting from a cryptographic hash function, such as `MD5` or `SHA512`, on input data. -- **Inherent**: A qualifier applied to a seqcol attribute indicating that the attribute is part of the definition of the sequence collection and therefore contributes to its digest. - **Length**: The number of characters in a sequence. +- **Qualifier**: A reserved term used in the schema to indicate a quality of an attribute, such as whether it is required, collated, or inherent. Qualifiers are listed below. - **Seqcol algorithm**: The set of instructions used to compute a digest from a sequence collection. - **Seqcol API**: The set of endpoints defined in the *retrieval* and *comparison* components of the seqcol protocol. - **Seqcol digest**: A digest for a sequence collection, computed according to the seqcol algorithm. @@ -80,6 +81,13 @@ Some other examples of common use cases where the use of seqcol is beneficial in - **Sequence collection**: A representation of 1 or more sequences that is structured according to the sequence collection schema - **Sequence collection attribute**: A property or feature of a sequence collection (*e.g.* names, lengths, sequences, or topologies). +### Attribute qualifiers + +- **Collated**: A qualifier applied to a seqcol attribute indicating that the values of the attribute match 1-to-1 with the sequences in the collection and are represented in the same order. +- **Inherent**: A qualifier applied to a seqcol attribute indicating that the attribute is part of the definition of the sequence collection and therefore contributes to its digest. +- **Passthru**: A qualifier applied to a seqcol attribute indicating that the attribute is *not digested* in transition from level 2 to level 1. So its value on level 1 representation the same as the level 2 representation. +- **Transient**: A qualifier applied to a seqcol attribute indicating that the attribute *cannot be retrieved through the `/attribute` endpoint*. + ## Seqcol protocol functionality The seqcol algorithm is based on the refget sequence algorithm for individual sequences and should use refget sequence servers to store the actual sequence data. @@ -90,13 +98,14 @@ The seqcol protocol defines the following: 1. *Schema* - The way an implementation should define the attributes of sequence collections it holds. 2. *Encoding* - An algorithm for computing a digest given a sequence collection. -3. *API* - A server RESTful API specification for retrieving and comparing sequence collections. +3. *API* - A server API specification for retrieving and comparing sequence collections. 4. *Ancillary attribute management* - A specification for organizing non-inherent metadata as part of a sequence collection. ### 1. Schema: Defining the attributes in the collection The first step for a Sequence Collections implementation is to define the *list of contents*, that is, what attributes are allowed in each collection, and which of these affect the digest. -The sequence collections standard is flexible with respect to the schema used, so implementations of the standard can use the standard with different schemas, as required by a particular use case. This divides the choice of content from the choice of algorithm, allowing the algorithm to be consistent even in situations where the content is not. +The sequence collections standard is flexible with respect to the schema used, so implementations of the standard can use the standard with different schemas, as required by a particular use case. +This divides the choice of content from the choice of algorithm, allowing the algorithm to be consistent even in situations where the content is not. This is an example of a general, minimal schema: @@ -131,16 +140,18 @@ properties: required: - names - lengths -inherent: - - lengths - - names - sequences +ga4gh: + inherent: + - names + - sequences ``` -This example schema is the minimal standard schema. We RECOMMEND that all implementations use this as a base schema, adding additional attributes as needed, but *without changing the inherent attributes list*, because this will keep the identifiers compatible across implementations. +This example schema is the minimal standard schema. Sequence collection objects that follow this basic minimal structure are said to be the *canonical seqcol object representation*. -Adding custom attributes to this schema will not break interoperability. -Nevertheless, extending this schema is only RECOMMENDED; implementations are still compliant if using custom schemas with custom inherent attributes. +We RECOMMEND that all implementations use this as a base schema, adding additional attributes as needed, which will not break interoperability. +We RECOMMEND *not changing the inherent attributes list*, because this will keep the identifiers compatible across implementations. +Implementations that use different inherent attributes are still compliant with the specification generally, but do so at the cost of top-level digest interoperability. For more information about community-driven updates to the standard schema, see [*Footnote F8*](#f8-adding-new-schema-attributes). @@ -194,10 +205,8 @@ The implementation `MUST` define its structure in a JSON Schema, such as the exa Implementations `MAY` choose to extend this schema by adding additional attributes. Implementations `MAY` also use a schema, but we `RECOMMEND` the schema extend the base schema defined above. This schema extends vanilla JSON Schema in two ways; first, it provides the `collated` qualifier. -For further details about the rationale behind collated attributes, see [*Footnote F2*](#f2-collated-attributes). Second, it specifies the `inherent` qualifier. -For further details about the rationale and examples of non-inherent attributes, see [*Footnote F3*](#f3-details-of-inherent-and-non-inherent-attributes). -Finally, another detail that may be unintuitive at first is that in the minimal schema, the `sequences` attribute is optional; for an explanation of why, see [*Footnote F4*](#f4-sequence-collections-without-sequences). +For further details about attribute qualifiers, see [*Section 4*](#4-extending-the-schema-schema-attribute-qualifiers). ##### Filter non-inherent attributes @@ -230,7 +239,7 @@ b'["SQ.2648ae1bacce4ec4b6cf337dcae37816","SQ.907112d17fcb73bcab1ed1c72b97ce68"," _* The above Python function suffices if (1) attribute keys are restricted to ASCII, (2) there are no floating point values, and (3) for all integer values `i`: `-2**63 < i < 2**63`_ - Also, notice that in this process, RFC-8785 is applied only to objects; we assume the sequence digests are computed through an external process (the refget sequences protocol), and are not computed as part of the sequence collection. The refget sequences protocol digests sequence strings without JSON-canonicalization. For more details, see [*Footnote F5*](#f5-rfc-8785-does-not-apply-to-refget-sequences). +Also, notice that in this process, RFC-8785 is applied only to objects; we assume the sequence digests are computed through an external process (the refget sequences protocol), and are not computed as part of the sequence collection. The refget sequences protocol digests sequence strings without JSON-canonicalization. For more details, see [*Footnote F5*](#f5-rfc-8785-does-not-apply-to-refget-sequences). #### Step 3: Digest each canonicalized attribute value using the GA4GH digest algorithm. @@ -266,6 +275,11 @@ wqet7IWbw2j2lmGuoKCaFlYS_R7szczz ``` +#### Exception for passthru attributes + +The above canonicalization/digesting procedure is applied by default to all attributes of a sequence collection; however, there can be some exceptions. +Any attribute qualified in the schema as a *passthru* attribute is NOT digested in this way. + #### Terminology Because the encoding algorithm is recursive, this leads to a few different ways to represent a sequence collection. We refer to these representations in "levels". The level number represents the number of "lookups" you'd have to do from the "top level" digest. So, we have: @@ -313,7 +327,7 @@ What you'd get with **2 database lookups** (1 recursive call). This is the most ``` -### 3. API: A server RESTful API specification for retrieving and comparing sequence collections. +### 3. API: A server API specification for retrieving and comparing sequence collections. The API has these top-level endpoints: @@ -362,10 +376,11 @@ properties: required: - names - lengths -inherent: - - lengths - - names - sequences +ga4gh: + inherent: + - sequences + - names ``` @@ -490,6 +505,12 @@ Example `/attribute/collection/names/:digest` return value: ["A","B","C"] ``` +The attribute endpoint MUST be functional for any attribute defined in the schema, *except those marked as transient or passthru*. +The endpoint MAY respond to requests for attributes marked as *passthru*. +The endpoint SHOULD NOT respond to requests for attributes marked as *transient*. +For more information on transient and passthru attributes, see [Section 4](#4-extending-the-schema-schema-attribute-qualifiers). + + ##### Definition of `object_type` The `/list` and `/attribute` endpoints both use an `:object_type` path parameter. The `object_type` should always be the *singular* descriptor of objects provided by the server. In this version of the Sequence Collections specification, the `object_type` is always `collection`; so the only allowable endpoints would be `/list/collection` and `/attribute/collection/:attribute_name/:digest`. We call this `object_type` because future versions of the specification may allow retrieving lists or attributes of other types of objects. @@ -499,8 +520,98 @@ The `/list` and `/attribute` endpoints both use an `:object_type` path parameter In addition to the primary top-level endpoints, it is RECOMMENDED that the service provide `/openapi.json`, an OpenAPI-compatible description of the endpoints. ---- -### 4. Ancillary attribute management: recommended non-inherent attributes + + + +### 4. Extending the schema: Schema attribute qualifiers + +#### 4.1 Introduction to attribute qualifiers + +The Sequence Collections specification is designed to be extensible. +This will let you build additional capability on top of the minimal Sequence Collections standard. +You can do this by extending the schema to include ancillary custom attributes. +To allow other services to understand something about what these attributes are for, you can annotate them in the schema using *attribute qualifiers*. +This allows you to indicate what *type* of attribute your custom attributes are, which govern how the service should respond to requests. +This section will describe the 4 attribute qualifiers you may add to the schema to qualify custom attributes. + +#### 4.2 Collated attributes + +Collated attributes are attributes whose values match 1-to-1 with the sequences in the collection and are represented in the same order. +A collated attribute by definition has the same number of elements as the number of sequences in the collection. +It is also in the same order as the sequences in the collection. + + +#### 4.3 Inherent attributes + +Inherent attributes are those that contribute to the digest. +The specification in section 1, *Encoding*, described how to structure a sequence collection and then apply an algorithm to compute a digest for it. +What if you have ancillary information that goes with a collection, but shouldn't contribute to the digest? +We have found a lot of useful use cases for information that should go along with a seqcol, but should not contribute to the *identity* of that seqcol. +This is a useful construct as it allows us to include information in a collection that does not affect the digest that is computed for that collection. +One simple example is the "author" or "uploader" of a reference sequence; this is useful information to store alongside this collection, but we wouldn't want the same collection with two different authors to have a different digest! Seqcol refers to these as *non-inherent attributes*, meaning they are not part of the core identity of the sequence collection. +Non-inherent attributes are defined in the seqcol schema, but excluded from the `inherent` list. + +See: [ADR on 2023-03-22 regarding inherent attributes](decision_record.md#2023-03-22-seqcol-schemas-must-specify-inherent-attributes) + +#### 4.4 Passthru attributes + +Passthru attributes are *not digested* in transition from level 2 to level 1. +In other words, the value of a passthru attribute is the same in the level 1 and level 2 representations +This is not the case for most attributes; most attributes of the canonical (level 2) seqcol representation are digested to create the level 1 representation. +But sometimes, we have an attribute for which digesting makes little sense. +These attributes are passed through the transformation, so they show up on the level 1 representation in the same form as the level 2 representation. +Thus, we refer to them as passthru attributes. + + +#### 4.5 Transient attributes + +Transient attributes are those that *cannot be retrieved* through the `/attribute` endpoint. +Most attributes of the sequence collection can be retrieved through the `/attribute` endpoint. +However, some attributes may not be retrievable. +For example, this could happen for an attribute that we intend to be used primarily as an identifier. +In this case, we don't necessarily want to store the original content that went into the digest into the database, because it might be redundant or whatever. +We really just want the final attribute. +These attributes are called transient because the content of the attribute is no longer stored and is therefore no longer retrievable. + +#### 4.6 Method of specifying attribute qualifiers + +In JSON Schema, there are 2 ways to qualify properties: 1) a local qualifier, using a key under a property; or 2) an object-level qualifier, which is specified with a keyed list of properties up one level. +For example, you annotate a property's `type` with a local qualifier, underneath the property, like this: + +```console +properties: + names: + type: array +``` + +However, you specify that a property is `required` by adding it to an object-level `required` list that's parallel to the `properties` keyword: + +```console +properties: + names: + type: array +required: + - names +``` + +In sequence collections, we define `collated` as a local qualifier. +Local qualifiers fit better for qualifiers independent of the object as a whole. +They are qualities of a property that persist if the property were moved onto a different object. +For example, the `type` of an attribute is consistent, regardless of what object that attribute were defined on. +In contrast, object-level qualifier lists fit better for qualifiers that depend on the object as a whole. +They are qualities of a property that depend on the object context in which the property is defined. +For example, the `required` modifier is not really meaningful except in the context of the object as a whole. +A particular property could be required for one object type, but not for another, and it's really the object that induces the requirement, not the property itself. + + +We reasoned that `inherent`, `transient`, and `passthru` are global qualifiers, like `required`, which describe the role of an attribute in the context of the whole object. +For example, an attribute that is inherent to one type of object need not be inherent to another. +Therefore, it makes sense to treat this concept the same way JSON schema treats `required`. +In contrast, the idea of `collated` describes a property independently: Whether an attribute is collated is part of the definition of the attribute; if the attribute were moved to a different object, it would still be collated. + +Finally, the 3 global qualiers are grouped under the 'ga4gh' key for consistency with other GA4GH specifications, and to group the seqcol-specific extended functionality into one place. + +### 5. Ancillary attribute management: recommended non-inherent attributes In *Section 1: Encoding*, we distinguished between *inherent* and *non-inherent* attributes. Non-inherent attributes provide a standardized way for implementations to store and serve additional, third-party attributes that do not contribute to the digest. @@ -508,12 +619,28 @@ As long as separate implementations keep such information in non-inherent attrib Furthermore, the structure for how such non-inherent metadata is retrieved will be standardized. Here, we specify standardized, useful non-inherent attributes that we recommend. -#### 4.1 The `sorted_name_length_pairs` attribute (`RECOMMENDED`) +#### 5.1 The `name_length_pairs` attribute (`RECOMMENDED`) + +The `name_length_pairs` attribute is a *non-inherent* attribute of a sequence collection with a formal definition, provided here. +This attribute provides a way to look up the ordered coordinate system (the "chrom sizes") for a sequence collection. +It is created deterministically from the `names` and `lengths` attributes in the collection; it *does not* depend on the actual sequence content, so it is consistent across two collections with different sequence content if they have the same `names` and `lengths`, which are correctly collated. +This attribute is `RECOMMENDED` to allow retrieval of the coordinate system for a given reference sequence collections. -The `sorted_name_length_pairs` attribute is a *non-inherent* attribute of a sequence collection with a formal definition, provided here. +Algorithm: + +1. Lump together each name-length pair from the primary collated `names` and `lengths` into an object, like `{"length":123,"name":"chr1"}`. +2. Build a collated list, corresponing to the names and lengths of the object (*e.g.* `[{"length":123,"name":"chr1"},{"length":456,"name":"chr2"}],...`) +3. Add as a collated attribute to the sequence collection object. + +The `name_length_pairs` attribute is *not inherent*, *not passthru*, and *not transient*. + +#### 5.1 The `sorted_name_length_pairs` attribute (`RECOMMENDED`) + +The `sorted_name_length_pairs` attribute is similar to the `name_length_pairs` attribute, but it is sorted. When digested, this attribute provides a digest for an order-invariant coordinate system for a sequence collection. Because it is *non-inherent*, it does not affect the identity (digest) of the collection. -It is created deterministically from the `names` and `lengths` attributes in the collection; it *does not* depend on the actual sequence content, so it is consistent across two collections with different sequence content if they have the same `names` and `lengths`, which are correctly collated, but with pairs not necessarily in the same order. +but with pairs not necessarily in the same order. + This attribute is `RECOMMENDED` to allow unified genome browser visualization of data defined on different reference sequence collections. For more rationale and use cases of `sorted_name_length_pairs`, see [*Footnote F7*](#f7-use-cases-for-the-sorted_name_length_pairs-non-inherent-attribute). Algorithm: @@ -522,9 +649,11 @@ Algorithm: 2. Canonicalize JSON according to the seqcol spec (using RFC-8785). 3. Digest each name-length pair string individually. 4. Sort the digests lexicographically. -5. Add as a non-inherent, non-collated attribute to the sequence collection object. +5. Add to the sequence collection object. -#### 4.2 The `sorted_sequences` attribute (`OPTIONAL`) +The `sorted_name_length_pairs` attribute is: non-inherent, non-collated, non-passthru, and transient. + +#### 5.3 The `sorted_sequences` attribute (`OPTIONAL`) The `sorted_sequences` attribute is a *non-inherent* attribute of a sequence collection, with a formal definition. Providing this attribute is `OPTIONAL`. @@ -541,6 +670,10 @@ Algorithm: 2. Canonicalize the resulting array (using RFC-8785). 3. Add to the sequence collection object as the `sorted_sequences` attribute, which is non-inherent and non-collated. + + +--- + ## Footnotes ### F1. Why use an array-oriented structure instead of a sequence-oriented structure? @@ -557,51 +690,6 @@ While the latter is intuitive, as it captures each sequence object with some acc See [ADR on 2021-06-30 on array-oriented structure](decision_record.md#2021-06-30-use-array-based-data-structure-and-multi-tiered-digests) - -### F2. Collated attributes - -In JSON Schema, there are 2 ways to qualify properties: 1) a local qualifier, using a key under a property; or 2) an object-level qualifier, which is specified with a keyed list of properties up one level. -For example, you annotate a property's `type` with a local qualifier, underneath the property, like this: - -```console -properties: - names: - type: array -``` - -However, you specify that a property is `required` by adding it to an object-level `required` list that's parallel to the `properties` keyword: - -```console -properties: - names: - type: array -required: - - names -``` - -In sequence collections, we chose to define `collated` as a local qualifier. Local qualifiers fit better for qualifiers independent of the object as a whole. -They are qualities of a property that persist if the property were moved onto a different object. -For example, the `type` of an attribute is consistent, regardless of what object that attribute were defined on. -In contrast, object-level qualifier lists fit better for qualifiers that depend on the object as a whole. -They are qualities of a property that depend on the object context in which the property is defined. -For example, the `required` modifier is not really meaningful except in the context of the object as a whole. A particular property could be required for one object type, but not for another, and it's really the object that induces the requirement, not the property itself. - -We reasoned that `inherent`, like `required`, describes the role of an attribute in the context of the whole object; an attribute that is inherent to one type of object need not be inherent to another. -Therefore, it makes sense to treat this concept the same way JSON schema treats `required`. -In contrast, the idea of `collated` describes a property independently: Whether an attribute is collated is part of the definition of the attribute; if the attribute were moved to a different object, it would still be collated. - - -### F3. Details of inherent and non-inherent attributes - -The specification in section 1, *Encoding*, described how to structure a sequence collection and then apply an algorithm to compute a digest for it. -What if you have ancillary information that goes with a collection, but shouldn't contribute to the digest? -We have found a lot of useful use cases for information that should go along with a seqcol, but should not contribute to the *identity* of that seqcol. -This is a useful construct as it allows us to include information in a collection that does not affect the digest that is computed for that collection. -One simple example is the "author" or "uploader" of a reference sequence; this is useful information to store alongside this collection, but we wouldn't want the same collection with two different authors to have a different digest! Seqcol refers to these as *non-inherent attributes*, meaning they are not part of the core identity of the sequence collection. -Non-inherent attributes are defined in the seqcol schema, but excluded from the `inherent` list. - -See: [ADR on 2023-03-22 regarding inherent attributes](decision_record.md#2023-03-22-seqcol-schemas-must-specify-inherent-attributes) - ### F4. Sequence collections without sequences Typically, we think of a sequence collection as consisting of real sequences, but in fact, sequence collections can also be used to specify collections where the actual sequence content is irrelevant. From 54049bbf19c0b4ab8543c15572597ef39c592da1 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 09:09:29 -0500 Subject: [PATCH 3/9] add details to new attribute qualifiers --- docs/seqcol.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/docs/seqcol.md b/docs/seqcol.md index 64d6eeb..0048f08 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -203,7 +203,7 @@ The object is a series of arrays with matching length (`3`), with the correspond For the rationale why this structure was chosen instead of an array of annotated sequences, see [*Footnote F1*](#f1-why-use-an-array-oriented-structure-instead-of-a-sequence-oriented-structure). The implementation `MUST` define its structure in a JSON Schema, such as the example schema defined in step 1. Implementations `MAY` choose to extend this schema by adding additional attributes. -Implementations `MAY` also use a schema, but we `RECOMMEND` the schema extend the base schema defined above. +Implementations `MAY` also use a different schema, but we `RECOMMEND` the schema extend the base schema defined above. This schema extends vanilla JSON Schema in two ways; first, it provides the `collated` qualifier. Second, it specifies the `inherent` qualifier. For further details about attribute qualifiers, see [*Section 4*](#4-extending-the-schema-schema-attribute-qualifiers). @@ -485,6 +485,8 @@ Example return value: } ``` +The `list` endpoint MUST be implemented, and MUST allow filtering any attribute defined in the schema, *except attributes marked as passthru*. +For attributes marked as *passthru*, the list endpoint MAY provide filtering capability, but the spec is silent on this behavior because passthru attributes may be of types other than string. #### 3.5 Attribute From e92a86b4600378dac2e43138beaaacfb1f43e18b Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 12:27:23 -0500 Subject: [PATCH 4/9] some renaming --- docs/seqcol.md | 27 ++++++++++++++++++++------- 1 file changed, 20 insertions(+), 7 deletions(-) diff --git a/docs/seqcol.md b/docs/seqcol.md index 0048f08..490c3cf 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -465,9 +465,11 @@ For more details about how to interpret the results of the comparison function t #### 3.4 List -- *Endpoint*: `GET /list/:object_type?page=:page&page_size=:page_size&:attribute1=:attribute_digest1&attribute2=:attribute_digest2` (`REQUIRED`) -- *Description*: Lists identifiers for a given object type in singular form (*e.g.* `/list/collection`). This endpoint provides a way to discover what sequence collections a service provides. Returned lists can be filtered to only objects with certain attribute values using query parameters. Page numbering begins at page 0 (the first page is page 0). -- *Return value*: The output is a paged list of identifiers following the GA4GH paging guide format, grouped into a `results` and a `pagination` section. If no `?:attribute=:attribute_value` query parameters are provided, the endpoint will return all items (paged). Adding one or more `:attribute` and `:attribute_digest` values as *query parameters* will filter results to only the collections with the given attribute digest. If multiple attributes are provided, the filter should require ALL of these attributes to match (so multiple attributes are treated with an `AND` operator). +- *Endpoint*: `GET /list/:object_type?page=:page&page_size=:page_size&:attribute1=:attribute1_level1_repr&attribute2=:attribute2_level1_repr` (`REQUIRED`) +- *Description*: Lists identifiers for a given object type in singular form (*e.g.* `/list/collection`). This endpoint provides a way to discover what sequence collections a service provides. + Returned lists can be filtered to only objects with certain attribute values using query parameters. + Page numbering begins at page 0 (the first page is page 0). +- *Return value*: The output is a paged list of identifiers following the GA4GH paging guide format, grouped into a `results` and a `pagination` section. If no `?:attribute=:attribute_level1_repr` query parameters are provided, the endpoint will return all items (paged). Adding one or more `:attribute` and `:attribute_digest` values as *query parameters* will filter results to only the collections with the given attribute digest. If multiple attributes are provided, the filter should require ALL of these attributes to match (so multiple attributes are treated with an `AND` operator). Example return value: @@ -485,8 +487,8 @@ Example return value: } ``` -The `list` endpoint MUST be implemented, and MUST allow filtering any attribute defined in the schema, *except attributes marked as passthru*. -For attributes marked as *passthru*, the list endpoint MAY provide filtering capability, but the spec is silent on this behavior because passthru attributes may be of types other than string. +The `list` endpoint MUST be implemented, and MUST allow filtering using any attribute defined in the schema, *except attributes marked as passthru*. +For attributes marked as *passthru*, the list endpoint MAY provide filtering capability. #### 3.5 Attribute @@ -508,8 +510,7 @@ Example `/attribute/collection/names/:digest` return value: ``` The attribute endpoint MUST be functional for any attribute defined in the schema, *except those marked as transient or passthru*. -The endpoint MAY respond to requests for attributes marked as *passthru*. -The endpoint SHOULD NOT respond to requests for attributes marked as *transient*. +The endpoint SHOULD NOT respond to requests for attributes marked as *passthru* OR *transient*. For more information on transient and passthru attributes, see [Section 4](#4-extending-the-schema-schema-attribute-qualifiers). @@ -613,6 +614,18 @@ In contrast, the idea of `collated` describes a property independently: Whether Finally, the 3 global qualiers are grouped under the 'ga4gh' key for consistency with other GA4GH specifications, and to group the seqcol-specific extended functionality into one place. + + +The qualifiers are all about transitions from different representations. + +qualifer | level1 | level2 +*NONE* | digest | main representation +passthru | yes | same as level1 +transient | yes | N/A (not present) +inherent + + + ### 5. Ancillary attribute management: recommended non-inherent attributes In *Section 1: Encoding*, we distinguished between *inherent* and *non-inherent* attributes. From 9e5664deb653bbc1637b832ec0ebf6bb9304b675 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 12:47:40 -0500 Subject: [PATCH 5/9] add endpoint behavior for qualified attributes --- docs/seqcol.md | 57 +++++++++++++++++++++++++++++++++----------------- 1 file changed, 38 insertions(+), 19 deletions(-) diff --git a/docs/seqcol.md b/docs/seqcol.md index 490c3cf..ac0c3b3 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -558,25 +558,54 @@ See: [ADR on 2023-03-22 regarding inherent attributes](decision_record.md#2023-0 #### 4.4 Passthru attributes -Passthru attributes are *not digested* in transition from level 2 to level 1. -In other words, the value of a passthru attribute is the same in the level 1 and level 2 representations +Passthru attributes have the same representation at level 1 and level 2. +In other words, they are *not digested* in transition from level 2 to level 1. This is not the case for most attributes; most attributes of the canonical (level 2) seqcol representation are digested to create the level 1 representation. But sometimes, we have an attribute for which digesting makes little sense. -These attributes are passed through the transformation, so they show up on the level 1 representation in the same form as the level 2 representation. +These attributes are passed through without transformation, so they show up on the level 1 representation in the same form as the level 2 representation. Thus, we refer to them as passthru attributes. +Here's how passthru attributes behave in the endpoints: +- `/list`: The server MAY allow filtering on passthru attributes, but this is not required. +- `/collection`: At both level 1 and level 2, the collection object includes the same passthru attribute representation. +- `/comparison`: Passthru attributes are listed in the 'attributes' section, but are not listed under 'array_elements'. +- `/attribute`: Passthru attributes cannot be used with the attribute endpoint, as the return would be the same as the query. + #### 4.5 Transient attributes -Transient attributes are those that *cannot be retrieved* through the `/attribute` endpoint. -Most attributes of the sequence collection can be retrieved through the `/attribute` endpoint. -However, some attributes may not be retrievable. -For example, this could happen for an attribute that we intend to be used primarily as an identifier. -In this case, we don't necessarily want to store the original content that went into the digest into the database, because it might be redundant or whatever. +A transient attribute is an attribute that only has a level 1 representation stored in the server. +Transient attributes therefore *cannot be retrieved* through the `/attribute` endpoint. +All other attributes of the sequence collection can be retrieved through the `/attribute` endpoint. +The transient qualifier would apply to attribute that we intend to be used primarily as an identifier. +In this case, we don't necessarily want to store the original content that went into the digest into the database. We really just want the final attribute. These attributes are called transient because the content of the attribute is no longer stored and is therefore no longer retrievable. -#### 4.6 Method of specifying attribute qualifiers +Here's how transient attributes behave in the endpoints: +- `/list`: No change; a transient attribute level1 representation can be used to list sequence collections that contain it. +- `/collection`: For level 1 representation, no change; the collection object includes the transient attribute level 1 representation. For level 2 representation, there *is* a change; transient attributes have no level 2 representation on the server, so the sequence collection SHOULD leave this attribute out of the level 2 representation. +- `/comparison`: Transient attributes are listed in the 'attributes' section, but are not listed under 'array_elements' because there is no level 2 representation. +- `/attribute`: Transient attributes cannot be used with the attribute endpoint (there is no value to retrieve) + + +#### 4.6 Qualifier summary table + +The global qualifiers are all concerned with how the representations are treated when converting between different detail levels. +The *inherent* qualifier is related to the level 1 → 0 transition. +It is true if the level 1 representation is included during creation of level 0 representation. +Then the *passthru* and *transient* qualifiers are related to the level 2 → 1 transition. + + +Qualifer | Level1? | Level2? | Notes +--- | ----- | ----- | ----- +*none* | as normal | full content | Default state; the level 1 representation is a digest of the level 2 representation. +passthru | as normal | same as level1 | True if the level 2 representation is the same as the level 1 representation. +transient | as normal | not present | True if the level 2 representation is not present. + + + +#### 4.7 Method of specifying attribute qualifiers In JSON Schema, there are 2 ways to qualify properties: 1) a local qualifier, using a key under a property; or 2) an object-level qualifier, which is specified with a keyed list of properties up one level. For example, you annotate a property's `type` with a local qualifier, underneath the property, like this: @@ -616,16 +645,6 @@ Finally, the 3 global qualiers are grouped under the 'ga4gh' key for consistency -The qualifiers are all about transitions from different representations. - -qualifer | level1 | level2 -*NONE* | digest | main representation -passthru | yes | same as level1 -transient | yes | N/A (not present) -inherent - - - ### 5. Ancillary attribute management: recommended non-inherent attributes In *Section 1: Encoding*, we distinguished between *inherent* and *non-inherent* attributes. From cf90134692b4ef680483f8e8b55d7201ce6c0f42 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 12:51:04 -0500 Subject: [PATCH 6/9] add qualifiers to custom attributes --- docs/seqcol.md | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-) diff --git a/docs/seqcol.md b/docs/seqcol.md index ac0c3b3..9c1fc95 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -660,13 +660,19 @@ This attribute provides a way to look up the ordered coordinate system (the "chr It is created deterministically from the `names` and `lengths` attributes in the collection; it *does not* depend on the actual sequence content, so it is consistent across two collections with different sequence content if they have the same `names` and `lengths`, which are correctly collated. This attribute is `RECOMMENDED` to allow retrieval of the coordinate system for a given reference sequence collections. -Algorithm: +##### Algorithm 1. Lump together each name-length pair from the primary collated `names` and `lengths` into an object, like `{"length":123,"name":"chr1"}`. 2. Build a collated list, corresponing to the names and lengths of the object (*e.g.* `[{"length":123,"name":"chr1"},{"length":456,"name":"chr2"}],...`) 3. Add as a collated attribute to the sequence collection object. -The `name_length_pairs` attribute is *not inherent*, *not passthru*, and *not transient*. +##### Qualifiers + +- inherent: false +- collated: false +- passthru: false +- transient: false + #### 5.1 The `sorted_name_length_pairs` attribute (`RECOMMENDED`) @@ -677,7 +683,7 @@ but with pairs not necessarily in the same order. This attribute is `RECOMMENDED` to allow unified genome browser visualization of data defined on different reference sequence collections. For more rationale and use cases of `sorted_name_length_pairs`, see [*Footnote F7*](#f7-use-cases-for-the-sorted_name_length_pairs-non-inherent-attribute). -Algorithm: +##### Algorithm 1. Lump together each name-length pair from the primary collated `names` and `lengths` into an object, like `{"length":123,"name":"chr1"}`. 2. Canonicalize JSON according to the seqcol spec (using RFC-8785). @@ -685,7 +691,12 @@ Algorithm: 4. Sort the digests lexicographically. 5. Add to the sequence collection object. -The `sorted_name_length_pairs` attribute is: non-inherent, non-collated, non-passthru, and transient. +##### Qualifiers + +- inherent: false +- collated: false +- passthru: false +- transient: **true** #### 5.3 The `sorted_sequences` attribute (`OPTIONAL`) @@ -698,13 +709,18 @@ Simply that for some large-scale use cases, comparing the sequence content witho In these cases, using the comparison function could be computationally prohibitive. This digest allows the comparison to be pre-computed, and more easily compared. -Algorithm: +##### Algorithm 1. Take the array of the `sequences` attribute (an array of sequence digests) and sort it lexicographically. 2. Canonicalize the resulting array (using RFC-8785). 3. Add to the sequence collection object as the `sorted_sequences` attribute, which is non-inherent and non-collated. +##### Qualifiers +- inherent: false +- collated: false +- passthru: false +- transient: false --- From 81d3791b0dda2084967e7a60bebf4562ef755127 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 12:52:31 -0500 Subject: [PATCH 7/9] name_length pairs is collated --- docs/seqcol.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/seqcol.md b/docs/seqcol.md index 9c1fc95..ea673bc 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -669,7 +669,7 @@ This attribute is `RECOMMENDED` to allow retrieval of the coordinate system for ##### Qualifiers - inherent: false -- collated: false +- collated: true - passthru: false - transient: false From 5e9d263fcc65dde29c60e6f2838a452b7738f8d6 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 12:53:07 -0500 Subject: [PATCH 8/9] make qualifiers recommended --- docs/seqcol.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/docs/seqcol.md b/docs/seqcol.md index ea673bc..43d01cb 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -666,7 +666,7 @@ This attribute is `RECOMMENDED` to allow retrieval of the coordinate system for 2. Build a collated list, corresponing to the names and lengths of the object (*e.g.* `[{"length":123,"name":"chr1"},{"length":456,"name":"chr2"}],...`) 3. Add as a collated attribute to the sequence collection object. -##### Qualifiers +##### Qualifiers (RECOMMENDED) - inherent: false - collated: true @@ -691,7 +691,7 @@ This attribute is `RECOMMENDED` to allow unified genome browser visualization of 4. Sort the digests lexicographically. 5. Add to the sequence collection object. -##### Qualifiers +##### Qualifiers (RECOMMENDED) - inherent: false - collated: false @@ -715,7 +715,7 @@ This digest allows the comparison to be pre-computed, and more easily compared. 2. Canonicalize the resulting array (using RFC-8785). 3. Add to the sequence collection object as the `sorted_sequences` attribute, which is non-inherent and non-collated. -##### Qualifiers +##### Qualifiers (RECOMMENDED) - inherent: false - collated: false From 20215caa96a17eecab8afeb7e22c92373fabebb6 Mon Sep 17 00:00:00 2001 From: nsheff Date: Wed, 20 Nov 2024 14:39:06 -0500 Subject: [PATCH 9/9] add decision on inherent schema location. See #84 --- docs/decision_record.md | 22 ++++++++++++ docs/seqcol.md | 73 ++++++++++++++-------------------------- docs/seqcol_rationale.md | 37 ++++++++++++++++++++ 3 files changed, 85 insertions(+), 47 deletions(-) diff --git a/docs/decision_record.md b/docs/decision_record.md index 1cbb184..b44d105 100644 --- a/docs/decision_record.md +++ b/docs/decision_record.md @@ -8,6 +8,23 @@ The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "S [TOC] +## 2024-11-20 Custom modifiers should live in the schema under the `ga4gh` key + +### Decision + +Any global custom modifiers should live under a `ga4gh` key in the schemea. Right now, this includes `inherent`, `transient`, and `passthru`. +Local modifiers (currently just `collated`) will continue to live, raw, under the attribute they describe. + + +### Rationale + +We want to follow the standard used in the other specs (VRS), and it also seems fine to have a place to lump together our custom modifiers. +We thought we could also do this for `collated`, as a local modifier, but opt not to right now because: there's only 1, it's a boolean, and it's not actually even used for anything in the spec at the moment, it is only there because it could be nice to use for a visualization of elements in a collection. The additional complexity of another layer just for this seems pointless at this point. + +### Linked issues + +- + ## 2024-11-13 Attributes can be designed as `passthru` or `transient`. ### Decision @@ -19,6 +36,11 @@ Transient attributes - Transient attributes are not retrievable from the attribute endpoint. Most attributes of the sequence collection can be retrieved through the /attribute endpoint. However, some attributes may not be retrievable. For example, this could happen for an attribute that we intend to be used primarily as an identifier. In this case, we don't necessarily want to store the original content that went into the digest into the database, because it might be redundant. We really just want the final attribute. These attributes are called transient because the content of the attribute is no longer stored and is therefore no longer retrievable. +Also, a few other related decisions we finalized: +- `collection` endpoint, level 2 collection representation should exclude transient attributes. +- `attribute` endpoint wouldn't provide anything for either transient or passthru attributes. +- Can passthru or transient attributes be inherent? They could, but it probably doesn't really make sense. Nevertheless, there's no reason to state that they cannot be. + ### Rationale As we worked on more advanced attributes, and with the addition of the `/attribute` endpoint, we realized these changes necessitate a bit more power for the schema to specify behavior of the attributes. For the basic seqcol attributes (names, lengths, sequences) and original endpoint, the general algorithm and basic qualifiers (required, inherent, collated) suffice to describe the representation. But some more nuanced attributes require additional qualifiers to describe their intention and how the server should be behave for the `/attribute` endpoint. For example, sorted_name_length_pairs and sorted_sequences are intended to provide alternative tailored identifiers and comparisons, and not necessarily useful for independent attribute lookup. Similarly, custom extra attributes, like author or alias, may be simple appendages that don't need the complex digesting procedure we use for the basic attributes. In order to flag such attributes in a way that can govern slightly different server expectations, we need a couple of additional advanced attribute qualifiers. For this purpose, we added the passthru and transient qualifiers. diff --git a/docs/seqcol.md b/docs/seqcol.md index 43d01cb..a3f2e73 100644 --- a/docs/seqcol.md +++ b/docs/seqcol.md @@ -31,7 +31,7 @@ In brief, the project specifies several procedures: 1. **An algorithm for encoding sequence collection identifiers.** The GA4GH standard [refget sequences](http://samtools.github.io/hts-specs/refget.html) specifies a way to compute deterministic sequence identifiers from individual sequences. Seqcol uses refget sequence identifiers and adds functionality to wrap them into collections of sequences. Seqcol also handles sequence attributes, such as their names, lengths, or topologies. Seqcol digests are defined by a hash algorithm, rather than an accession authority, and are thus decentralized and usable for private sequence collections, cases without connection to a central database, or validation of sequence collection content and provenance. 2. **An API describing lookup and comparison of sequence collections.** Seqcol specifies an http API to retrieve the sequence collection given a digest. A main use case is to reproduce the exact sequence collection (*e.g.* reference genome) used for analysis, instead of guessing based on a human-readable identifier. Seqcol also provides a standardized method of comparing the contents of two sequence collections. This comparison function can *e.g.* be used to determine if analysis results based on different references genomes are compatible. -3. **Recommended ancillary, non-inherent attributes.** Finally, the protocol defines several recommended procedures that will improve the compatibility across Seqcol servers, and beyond. +3. **Recommended ancillary attributes.** Finally, the protocol defines several recommended procedures that will improve the compatibility across Seqcol servers, and beyond. ## Use cases @@ -153,7 +153,7 @@ We RECOMMEND that all implementations use this as a base schema, adding addition We RECOMMEND *not changing the inherent attributes list*, because this will keep the identifiers compatible across implementations. Implementations that use different inherent attributes are still compliant with the specification generally, but do so at the cost of top-level digest interoperability. -For more information about community-driven updates to the standard schema, see [*Footnote F8*](#f8-adding-new-schema-attributes). +For more information about community-driven updates to the standard schema, see [*Footnote F5*](#f5-adding-new-schema-attributes). ### 2. Encoding: Computing sequence digests from sequence collections @@ -168,7 +168,10 @@ The steps of the encoding process are: - **Step 4**. Apply [RFC-8785 JSON Canonicalization Scheme](https://www.rfc-editor.org/rfc/rfc8785) again to canonicalize the JSON of the new seqcol object representation. - **Step 5**. Digest the final canonical representation again using the GA4GH digest algorithm. -Example Python code for computing a seqcol digest can be found in the [tutorial for computing seqcol digests](digest_from_collection.md). These steps are described in more detail below: +Example Python code for computing a seqcol digest can be found in the [tutorial for computing seqcol digests](digest_from_collection.md). +For information about the possibilty of deviating from this procedure for custom attributes, see [*Footnote F6*](#f6-custom-encoding-algorithms). + +These steps are described in more detail below: #### Step 1: Organize the sequence collection data into *canonical seqcol object representation*. @@ -239,12 +242,12 @@ b'["SQ.2648ae1bacce4ec4b6cf337dcae37816","SQ.907112d17fcb73bcab1ed1c72b97ce68"," _* The above Python function suffices if (1) attribute keys are restricted to ASCII, (2) there are no floating point values, and (3) for all integer values `i`: `-2**63 < i < 2**63`_ -Also, notice that in this process, RFC-8785 is applied only to objects; we assume the sequence digests are computed through an external process (the refget sequences protocol), and are not computed as part of the sequence collection. The refget sequences protocol digests sequence strings without JSON-canonicalization. For more details, see [*Footnote F5*](#f5-rfc-8785-does-not-apply-to-refget-sequences). +Also, notice that in this process, RFC-8785 is applied only to objects; we assume the sequence digests are computed through an external process (the refget sequences protocol), and are not computed as part of the sequence collection. The refget sequences protocol digests sequence strings without JSON-canonicalization. For more details, see [*Footnote F2*](#f2-rfc-8785-does-not-apply-to-refget-sequences). #### Step 3: Digest each canonicalized attribute value using the GA4GH digest algorithm. Apply the GA4GH digest algorithm to each attribute value. -The GA4GH digest algorithm is described in detail in [*Footnote F6*](#f6-the-ga4gh-digest-algorithm). +The GA4GH digest algorithm is described in detail in [*Footnote F3*](#f3-the-ga4gh-digest-algorithm). This converts the value of each attribute in the seqcol into a digest string. Applying this to each value will produce the following structure: @@ -645,7 +648,7 @@ Finally, the 3 global qualiers are grouped under the 'ga4gh' key for consistency -### 5. Ancillary attribute management: recommended non-inherent attributes +### 5. Recommended ancillary attributes In *Section 1: Encoding*, we distinguished between *inherent* and *non-inherent* attributes. Non-inherent attributes provide a standardized way for implementations to store and serve additional, third-party attributes that do not contribute to the digest. @@ -658,7 +661,7 @@ Here, we specify standardized, useful non-inherent attributes that we recommend. The `name_length_pairs` attribute is a *non-inherent* attribute of a sequence collection with a formal definition, provided here. This attribute provides a way to look up the ordered coordinate system (the "chrom sizes") for a sequence collection. It is created deterministically from the `names` and `lengths` attributes in the collection; it *does not* depend on the actual sequence content, so it is consistent across two collections with different sequence content if they have the same `names` and `lengths`, which are correctly collated. -This attribute is `RECOMMENDED` to allow retrieval of the coordinate system for a given reference sequence collections. +This attribute is `RECOMMENDED` to allow retrieval of the coordinate system for a given reference sequence collection. ##### Algorithm @@ -681,7 +684,7 @@ When digested, this attribute provides a digest for an order-invariant coordinat Because it is *non-inherent*, it does not affect the identity (digest) of the collection. but with pairs not necessarily in the same order. -This attribute is `RECOMMENDED` to allow unified genome browser visualization of data defined on different reference sequence collections. For more rationale and use cases of `sorted_name_length_pairs`, see [*Footnote F7*](#f7-use-cases-for-the-sorted_name_length_pairs-non-inherent-attribute). +This attribute is `RECOMMENDED` to allow unified genome browser visualization of data defined on different reference sequence collections. For more rationale and use cases of `sorted_name_length_pairs`, see [*Footnote F4*](#f4-use-cases-for-the-sorted_name_length_pairs-non-inherent-attribute). ##### Algorithm @@ -740,45 +743,11 @@ While the latter is intuitive, as it captures each sequence object with some acc See [ADR on 2021-06-30 on array-oriented structure](decision_record.md#2021-06-30-use-array-based-data-structure-and-multi-tiered-digests) -### F4. Sequence collections without sequences - -Typically, we think of a sequence collection as consisting of real sequences, but in fact, sequence collections can also be used to specify collections where the actual sequence content is irrelevant. -Since this concept can be a bit abstract for those not familiar, we'll try here to explain the rationale and benefit of this. -First, consider that in a sequence comparison, for some use cases, we may be primarily concerned only with the *length* of the sequence, and not the actual sequence of characters. -For example, BED files provide start and end coordinates of genomic regions of interest, which are defined on a particular sequence. -On the surface, it seems that two genomic regions are only comparable if they are defined on the same sequence. -However, this not *strictly* true; in fact, really, as long as the underlying sequences are homologous, and the position in one sequence references an equivalent position in the other, then it makes sense to compare the coordinates. -In other words, even if the underlying sequences aren't *exactly* the same, as long as they represent something equivalent, then the coordinates can be compared. -A prerequisite for this is that the *lengths* of the sequence must match; it wouldn't make sense to compare position 5,673 on a sequence of length 8,000 against the same position on a sequence of length 9,000 because those positions don't clearly represent the same thing; but if the sequences have the same length and represent a homology statement, then it may be meaningful to compare the positions. - -We realized that we could gain a lot of power from the seqcol comparison function by comparing just the name and length vectors, which typically correspond to a coordinate system. -Thus, actual sequence content is optional for sequence collections. -We still think it's correct to refer to a sequence-content-less sequence collection as a "sequence collection" -- because it is still an abstract concept that *is* representing a collection of sequences: we know their names, and their lengths, we just don't care about the actual characters in the sequence in this case. -Thus, we can think of these as a sequence collection without sequence characters. - -An example of a canonical representation (level 2) of a sequence collection with unspecified sequences would be: - -``` -{ - "lengths": [ - "1216", - "970", - "1788" - ], - "names": [ - "A", - "B", - "C" - ] -} -``` - -### F5. RFC-8785 does not apply to refget sequences +### F2. RFC-8785 does not apply to refget sequences A note to clarify potential confusion with RFC-8785. While the sequence collection specification determines that RFC-8785 will be used to canonicalize the JSON before digesting, this is specific to sequence collections, it *does not apply to the original refget sequences protocol*. According to the sequences protocol, sequences are digested as un-quoted strings. If RFC-8785 were applied at the level of individual sequences, they would be quoted to become valid JSON, which would change the digest. Since the sequences protocol predated the sequence collections protocol, it did not use RFC-8785; and anyway, the sequences are just primitive types so a canonicalization scheme doesn't add anything. This leads to the slight confusion that RFC-8785 canonicalization is only applied to the objects in the sequence collections, and not to the primitives when the underlying sequences are digested. - -### F6. The GA4GH digest algorithm +### F3. The GA4GH digest algorithm The GA4GH digest algorithm, `sha512t24u`, was created as part of the [Variation Representation Specification standard](https://vrs.ga4gh.org/en/stable/impl-guide/computed_identifiers.html). This procedure is described as ([Hart _et al_. 2020](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0239883)): @@ -803,7 +772,7 @@ def sha512t24u_digest(seq: bytes) -> str: See: [ADR from 2023-01-25 on digest algorithm](decision_record.md#2023-01-25-digest-algorithm) -### F7. Use cases for the `sorted_name_length_pairs` non-inherent attribute +### F4. Use cases for the `sorted_name_length_pairs` non-inherent attribute One motivation for this attribute comes from genome browsers, which may display genomic loci of interest (*e.g.* BED files). The genome browser should only show BED files if they annotate the same coordinate system as the reference genome. @@ -820,7 +789,7 @@ Thus, in a production setting, the full compatibility check can be reduced to a See: [ADR from 2023-07-12 on sorted name-length pairs](decision_record.md#2023-07-12-implementations-should-provide-sorted_name_length_pairs-and-comparison-endpoint) -### F8. Adding new schema attributes +### F5. Adding new schema attributes A strength of the seqcol standard is that the schema definition can be modified for particular use cases, for example, by adding new attributes into a sequence collection. This will allow different communities to use the standard without necessarily needing to subscribe to identical schemas, allowing the standard to be more generally useful. @@ -832,4 +801,14 @@ The goal is not to include all possible attributes in the schema, but just those An implementation may propose a new attribute to be added to this extended schema by raising an issue on the GitHub repository. The proposed attributes and definition can then be approved through discussion during the refget working group calls and ultimately added to the approved extended seqcol schema. These GitHub issues should be created with the label 'schema-term'. -You can follow these issues (or raise your own) at . \ No newline at end of file +You can follow these issues (or raise your own) at . + +### F6. Custom encoding algorithms + +A core part of Sequence Collections specification is the *encoding* algorithm, which describes how to create the digest for a sequence collection. +The encoding process can be divided into two steps; first, the attributes are encoded into the level 1 representation, and then this is encoded to produce the final digest (also called the level 0 or top level representation). +The first part of this process, encoding from level 2 to level 1, is the default; this is applied to any attributes that don't have something else defined specifically as part of the attribute definition. +This is the way all the minimal attributes (names, lengths, and sequences) should behave. +But custom attributes MAY diverge from this approach by defining their own encoding procedure that defines how the level 1 digest is computed from the level 2 representation. +For example, in the list of recommended ancillary attributes, `name_length_pairs` does not define a custom procedure for encoding, so this would follow the default procedure. +An alternative custom attribute, though, MAY specify how this encoding procedure happens. diff --git a/docs/seqcol_rationale.md b/docs/seqcol_rationale.md index f85d072..7797c4c 100644 --- a/docs/seqcol_rationale.md +++ b/docs/seqcol_rationale.md @@ -82,3 +82,40 @@ One final important point. Sometimes people seeing the standard for the first ti For reasons outlined in the specification, for the actual computing of the identifier, it's important to use the array-based structure -- this is what enables us to use the "level 1" digests for certain comparison questions, and also provides critical performance benefits for extremely large sequence collections. But don't let this dissuade you! My critical point is this: *the way to compute the interoperable identifier does not force you to structure your data in a certain way in your service* -- it's simply the way you structure the data when you compute its identifier. You are, of course, free to store a collection however you want, in whatever structure makes sense for you. You'd just need to structure it according to the standard if you wanted to implement the algorithm for computing the digest. In fact, my implementation provides a way to retrieve the collection information in either structure. + + + + + +### Sequence collections without sequences + +Typically, we think of a sequence collection as consisting of real sequences, but in fact, sequence collections can also be used to specify collections where the actual sequence content is irrelevant. +Since this concept can be a bit abstract for those not familiar, we'll try here to explain the rationale and benefit of this. +First, consider that in a sequence comparison, for some use cases, we may be primarily concerned only with the *length* of the sequence, and not the actual sequence of characters. +For example, BED files provide start and end coordinates of genomic regions of interest, which are defined on a particular sequence. +On the surface, it seems that two genomic regions are only comparable if they are defined on the same sequence. +However, this not *strictly* true; in fact, really, as long as the underlying sequences are homologous, and the position in one sequence references an equivalent position in the other, then it makes sense to compare the coordinates. +In other words, even if the underlying sequences aren't *exactly* the same, as long as they represent something equivalent, then the coordinates can be compared. +A prerequisite for this is that the *lengths* of the sequence must match; it wouldn't make sense to compare position 5,673 on a sequence of length 8,000 against the same position on a sequence of length 9,000 because those positions don't clearly represent the same thing; but if the sequences have the same length and represent a homology statement, then it may be meaningful to compare the positions. + +We realized that we could gain a lot of power from the seqcol comparison function by comparing just the name and length vectors, which typically correspond to a coordinate system. +Thus, actual sequence content is optional for sequence collections. +We still think it's correct to refer to a sequence-content-less sequence collection as a "sequence collection" -- because it is still an abstract concept that *is* representing a collection of sequences: we know their names, and their lengths, we just don't care about the actual characters in the sequence in this case. +Thus, we can think of these as a sequence collection without sequence characters. + +An example of a canonical representation (level 2) of a sequence collection with unspecified sequences would be: + +``` +{ + "lengths": [ + "1216", + "970", + "1788" + ], + "names": [ + "A", + "B", + "C" + ] +} +``` \ No newline at end of file