Relations in GO for 2009

Intro

• We have many relations ready to GO live in the scratch directory– within GO ontologies– across GO ontologies– between GO and external ontologies– Both cross product (N+S conditions) and regular links

• Requires a fundamental change in how we and our users think about GO and annotations– Tools that make use of these will better serve users

Relations in GO

• In the beginning there was is_a and part_of– Benefits: simplicity

• We could effectively ignore relations• Most tools and users effectively do this

– Speculation: recent introduction of regulates had no effect on majority of users

– Drawbacks: lack of expressivity• We need more relations

– Regulation– Spatial relations– has_part for Process-Function– annotations

Example of a relation rule in GO

• Rule:– A is_a B, B is_a C A is_a C

• Example:• We can generalize this by having a rule for transitive

relations– transitive r, A r B, B r C A r C

• We can also write this as a composition rule:– is_a . is_a is_a– Open question:

• does this notation help or hinder??

Transitivity

• We currently have two transitive relations in GO:– is_a . is_a is_a– part_of . part_of part_of

• Example:– mitotic prophase part_of mitosis– In GO, part_of is an all-some relation

• regulates is not defined to be transitive in GO• (but the majority of tools still treat it as if it were!)

• Example:

Composition with is_a• Any relation that follows the all-some pattern composes

with is_a to itself• Example:

– (all) nucleus part_of (some) cell• Composition:

– is_a . R R– R . is_a R

• Example:– (all) mitotic prophase part_of (some) mitosis– mitosis is_a cell cycle phase

• – (all) mitotic prophase part_of (some) cell cycle phase

is_a part_of

is_a is_a part_of

part_of part_of part_of

Read row first, the column(so far the table is symmetric)

Composition Table

is_a part_of

is_a is_a part_of

part_of part_of part_of

Composition Tablemitotic prophase part_of mitosis is_a cell cycle phase

(all) mitotic prophase part_of (some) cell cycle phase

is_a part_of

is_a is_a part_of

part_of part_of part_of

Chained compositionsA part_f B is_a C is_a D part_of E

A part_of B is_a D part_of E

A part_of D part_of E

A part_of E

order of reductiondoes not matter

regulates transitive_over part_of

• regulates . part_of regulates

inferredlink inferred link

regulates transitive_over part_of

• regulates . part_of regulates

inferredlink inferred link

(all) RoSPoMCCregulates (some) MCC

regulates transitive_over part_of

• regulates . part_of regulates

inferredlink inferred link

(all) RoSPoMCCregulates (some) MCC

is_a part_of regulates

is_a is_a part_of regulates

part_of part_of part_of -

regulates regulates regulates -

Composition Table: Regulates

is_a part_of regulates

is_a is_a part_of regulates

part_of part_of part_of -

regulates regulates regulates -

Composition Table: Regulates

regulates . part_of regulates

is_a part_of regulates

is_a is_a part_of regulates

part_of part_of part_of N/A

regulates regulates regulates -

Composition Table: Regulates

part_of . regulates N/A

is_a part_of regulates

is_a is_a part_of regulates

part_of part_of part_of -

regulates regulates regulates indirectly regulates

We have the option of defining additional relationsThese may be entirely implicit (i.e. we would never assert indirectly regulates in GO)

regulates . regulates indirectly regulates

is_a part_of regulates indirectly regulates

is_a is_a part_of regulates indirectly regulates

part_of part_of part_of - -

regulates regulates regulates indirectly regulates

indirectly regulates

indirectly regulates

indirectly regulates

indirectly regulates

indirectly regulates

indirectly regulatesRegulates is not transitive

Indirectly regulates is transitive

is_a part_of regulates indirectly regulates

is_a I P R ~R

part_of P P - -

regulates R R ~R ~R

indirectly regulates

~R ~R ~R ~R

USE SYMBOLS?OR IS THIS GETTING TOO ABSTRACT?

Sub-relations

• regulates– negatively_regulates– positively_regulates

is_a part_of regulates + regulates

- regulates

is_a I P R +R -R

part_of P P

regulates R R

+ regulates

+R +R +R

- regulates

-R -R -R

is_a part_of regulates + regulates

- regulates

indirectly regulates

is_a I P R +R -R ~R

part_of P P - - - -

regulates R R ~R ~R ~R ~R

+ regulates

+R +R +R ~+R ~-R ~R

- regulates

-R -R -R ~-R ~+R ~R

indirectly regulates

~R ~R ~R ~R ~R ~R

Sub-relations + indirect

RR

R+R+ R-R-

~R~R

~R+~R+ ~R-~R-

normal regulates relations asserted in GO

indirect regulates relations never asserted, only implied

Regulation relation lattice

RDRD

RD+RD+ RD-RD-

~R~R

~R+~R+ ~R-~R-

renamed to DIRECTLY regulates? indirect regulates relations never asserted, only implied

~RG~RG

~RG+~RG+ ~RG-~RG-

super-relation ofindirect and direct regulation(call this one “regulates”?)

has_part• NOT the inverse of part_of at the ontology level• Example:

– nucleus part_of cell: YES• every nucleus is part_of some cell

– by definition; e.g. extruded nuclei are ex-nuclei– cell has_part nucleus: NO

• not every cell has_part nucleus– mammalian erythrocytes, bacteria

• Example:– <pf example here>– <summarise pf progress>

Annotations and relations

• not just an ontology issue– this is of relevance to annotations too…

• The current simple methodology of propagating annotations up the graph only works for a small subset of relations– To understand how annotations and new relations

interact we must think in terms of gene product relations

Gene product relations

• What is the relation between a gene product and– A molecular function?– A biological process?– A cellular component?

• Why care?• What’s wrong with “annotated_to”?

– We need to define these relations:• to do justice to the biology• to be able to deal with new relations within the GO itself

Why we should care

• How should annotation queries, analysis tools (slimmers, enrichment tools) etc treat the (pseudo-)new regulates relation?

• How should we recommend the process-function links be vizualized?

• How should these links be treated in queries?

Proposed relations for gene products

• For MF and BP:– has_potential– has_function_during

• For CC:– localized_to

– This is more specific than has_location• A gene product may travel through different locations

– Formally:• GP localized_to CC : GP executes some function in CC

Names TBDMFs are ontologically like BPs(bfo processes)….

How to read a GAF

• <gene product> <rel> <GO term>• gene product may not be explicitly in GAF

– that’s OK– gene as proxy

• The relation does NOT apply to the gene however• genes are only localized_to chromosomes, and only

participate in gene expression. It’s the products that do the work

• <rel> is implicit, depending on F, C or P• Examples:

Annotation relation composition• is_a

– always propagate over is_a• localized_to . is_a localized_to• has_function_in . is_a has_function_in

• part_of• localized_to . part_of localized_to• has_function_in . part_of has_function_in

• This is effectively what we do with gene product annotations now

• post-hoc logical justification for why it’s OK to propagate

Annotation relation composition: regulates

• regulates– localized_to . regulates NEVER POSSIBLE

• localized_to never has a process as target• regulates always has process as subject

– has_function_in . regulates regulator_of

• This introduces an addition implicit relation that can be used to sum gene product results– Fake AmiGO screenshot here

Annotation relation composition: inter-ontology links

• We have 183 CC->MF/BP links in scratch• regulates

– localized_to . has_function_in ??may_contribute_to??• Example:

• RPS25A localized_to ribosome• ribosome has_function_in protein biosynthesis

– • RPS25A ??has_function_in?? protein biosynthesis

• No need for curator to make explicit IC claims

• Q: we never want “may” in relation names?• Can we make a stronger claim?• How does a curator know when to make an IC claim here?• Potential confusion with contributes_to qualifier

Annotation relations and has_part

• Need some graphical illustrations• See

– http://wiki.geneontology.org/index.php/Has_part

– for now

Qualifiers

• Annotation qualifiers (contributes_to) have the effect of modifying the relation– NOT is not a qualifier – it is a logical operator

• We can add new relations to the qualifier column– geneProductA acted_on_during protein secretion

by the type II secretion system

Secondary taxon IDs

Cell component relations

• We have 674 xp defs within CC in scratch– adjacent_to– surrounds/surrounded_by– spans– overlaps

• Use case: reactome

• Can we say anything about gene products here?– we can perform spatial gene product queries

Spatial reasoning

– spans . adjacent_to overlaps (??TBD!!)

– SUN-KASH complex spans nuclear inner membrance

– nuclear inner membrane adjacent_to nuclear lumen

– – SUN-KASH complex overlaps nuclear lumen

Links from BP to external ontologies

• Process-continuant links– A has_function_in cysteine biosynthesis

• A ??has_participant?? cysteine• this is true but can we make stronger claims

– A has_function_in heart development• A has_participant heart• c.f. heart process, TAZ gene

• How can we use this?– Browse GO annotations via other ontologies– Enrichment using anatomy terms…– AmiGO screenshots

what next?

Won’t this confuse users?

• We will provide a pre-made inferred relation table for all of GO– we could do this for gps too but it would be over a

billion entries..

• We can always distribute a dumbGO– just is_a and part_of, not even regulates

• Need more guidance on how this can be used

Discussion

What’s next?• Move relations into GO editors file

– post OE2– CC-self

• spatial relations– BP->MF

• has_part• regulates

– BP->BP• has_part (??)

– External onts• Dual releases? dumbGO and fullGO?• Fix GOC tools (AmiGO, slimmer, enrichment, graphviz, refG) to deal

appropriately– OE2 should already be fine

• Educate non-GOC folks