ICBO 2011 July 28-30, 2011 Buffalo, New York,

An Advanced Strategy for Integration of Biological Measurement Data

Hiroshi Masuya 1 , Georgios V. Gkoutos 2 , Nobuhiko Tanaka 1 , Kazunori Waki 1 , Yoshihiro Okuda 3 , Tatsuya Kushida 3 , Norio Kobayashi 4 , Koji Doi 4 , Kouji Kozaki 5 , Robert Hoehndorf 1 , Shigeharu Wakana 1 , Tetsuro Toyoda 4 Riichiro Mizoguchi 5 and 1: RIKEN BioResource Center, Tsukuba, Japan 2: Department of Genetics, University of Cambridge, UK 3: NalaPro Technologies, Inc, Tokyo, Japan 4: RIKEN BASE, Yokohama Japan 5: Department of Knowledge Systems, ISIR, Osaka University, Japan

Motivation of this study

Phenotypes represent a broad range of variations in measured qualities Sophisticated informatics infrastructure (ontology) Integrated phenotypic information whole

Organism A Organism B Organism C Organism D Mining… Biological knowledge

To contribute to development of the informatics infrastructure for the description, exchange and mining of phenotypic data.

Phenotypic Quality (PATO):

PATO provides a practical basis for vocabulary and semantics for the description of phenotype information across species.

•

Single hierarchy model of “quality” suite for BFO

•

Standard of phenotype annotation across species.

(“EQ” annotation)

•

Less confusions than “EAV” annotation for non-ontology-familiar people.

•

Basis of inferences of cross-species phenotype equivalence with EQ.

(e.g. mouse phenotype and disease) MP:0002269 ! muscular atrophy

E Q

MA:0000015 ! muscle PATO:0001623 ! atrophied FMA:30316 ! muscle

E

PATO:0001623 ! atrophied

Q

HP:0003202 ! Amyotrophy

Expansion of PATO

We attempted to expand the PATO ontology to ensure a more advanced, explicit and consistent knowledge framework.

Objectives: 1. To provide fundamental classification of quality values on the basis of measurement scales.

2. To provide strict data model to operate context dependencies of ordinal values.

3. To provide model of datum (or description) as a informational entity with the structure of common formalisms.

Fundamental classification of quality-value (1)

Refrain from 2-hiearchy model (and EAV formalism)

There were a lot of discussions for PATO to take 1 hiearchy and EQ…

1.

Number of studies claims that the fundamental classification of values: “scales of measurement” (Stevens S.S, 1946) is beneficial for data integration in the field of experimental science.

length

20cm Long, short

temperature

37 ℃ 310.15K

high, low

color

red, blue..

This classification takes as starting point the mathematical operation!

Fundamental classification of quality-value (2)

2.

Foundation of explicit description of change of quality is needed

Color 1: green to orange Color 2: orange to green 1 2 t1 1 2 t2

Ontology

BFO, PATO DOLCE Growing boy and his height quality

System of quality

1-hiearchy 2-hiearchy

Formalism

EQ EAV Explicit description of color change is needed.

Qualitative and quantitative descriptions are integrated in a single knowledge framework in DOLCE. For the coordination of ongoing efforts, equivalence mapping of these systems is beneficial.

Model of context-dependency of ordinal value (1)

I’m big!!

Problem of “ large ant and small elephant”

How to classify value instances?

Context A: simple comparison I’m small..

value A value C “Small” class value B value D “Large” class smaller Threshold X (some value) larger

Model of context-dependency of ordinal value (2) Problem of “ large ant and small elephant”

I’m big!!

How to classify value instances?

I’m small..

Context B: deviation based comparison (context of inference of cross-species equivalence of phenotypes)

value A value C deviation Threshold Y1 and Y2 (deviation-based value) value B value D “abnormally large” class larger smaller “normal size” class larger smaller “abnormally small” class

Knowledge model of context dependencies of ordinal scale values is needed!

Model of datum as an informational entity

Current version of DOLCE, BFO and PATO deal only with the primary reality and do not deal with quality description.

1. Distinction of a “true value” and an “empirical measurement” as an approximation is needed.

weigh t weigh t Reality (Unknown…) weigh t Information

2. Modeling of informational entities with common formalisms (eg. EQ, EAV and so on) and their relationships would be useful!

Reality (Unknown...) weigh t

EQ EAV

weigh t weigh t Information

BFO quality PATO OBI

Expansion of PATO with YAMATO framework

Mapping

YAMATO

Interoperability

DOLCE A reference ontology

“PATO2YAMATO”

•

Equivalence mapping between 1- and 2-hiearcy models

quale quality-space •

Model of context dependency

•

Model of datum with common formalisms

Practical use based proposals… • •

Yet Another More Advanced Top-Level Ontology

•

Features: (YAMATO: Mizoguchi, 2009) Framework of interoperability of quality-related concepts between top-level ontologies. Support of classification of scales of measurements.

Model of context dependency with “role” Detailed model of “representation” (an informational object) that involves quality representation.

Equivalence mapping of 1- and 2-hearcy model

BFO

(Upper level)

YAMATO

(Upper level)

quality quality_ quantity quality

(convertible) identical

property generic quality

identical

quality value

DOLCE

(Upper level)

region quality space quale

About 1,000 PATO terms were manually mapped to YAMATO framework. Classification of quality value (scales of measurements : Stevens S.S, 1946)

identical

Modeling of context dependency with “role”

I’m a teacher.

An entity often plays different “ roles ” with different characteristics under different contexts

(at school) (at home) I’m a husband In the

distribution for weight,

some

large-roles

thereby becomes

weight quality values

playing

role holders, abnormally heavy

context

role-holder

（ Entity playing a role)

Distribution for weight

role

Abnormally heavy

potential player

large-role weight quality value

depend on playable Concept model of role and role-holder heavier than normal value qualitative value for weight

Modeling of context dependency with “role”

I’m a teacher.

An entity often plays different “ roles ” with different characteristics under different contexts

(at school) (at home) I’m a husband In the

distribution for weight,

some

large-roles

thereby becomes

weight quality values

playing

role holders, abnormally heavy

context

Potential

player Role-holder Implementation and representation in Hozo ontology editor

Inter-relationships among contexts

Classification of organisms

Inherit Inherit

Inference of classification: ”Abnormally light in elephant is lighter than abnormally heavy in ant” “abnormally heavy”

Context of distribution of weight in elephant

Coordination of ordinary values under different contexts

Context of distribution of weight in ant

Abnormally heavy in elephant

larger

Normal weight in elephant

larger

Abnormally light in elephant

larger

Abnormally heavy in ant

larger

Normal weight in ant

larger

Abnormally light in ant

Simple comparison context

Quality representation in YAMATO

YAMATO provides “quality representation” for the foundation of formalized informational entities such as EQ, EAV and so on.

Reality Informational entities

Quality representation Weight Quality (Symbolization) Quality representation is modeled in the consistent way for content bearing informational entity, “representation”.

Basic structure for representation by symbol

quality representation

EP (=EQ) (BFO, PATO) EAV (DOLCE) Sentence of natural language Coding of genetic information Tupple *entity, #property Triple natural language alphabet nucleotide sequence molecular symbol (Mizoguchi, 2004) quality measurement *entity, #generic quality, value quality measurement anything… Specification of gene product

*: symbolization operation, #: Class => individual operation (equivalent with punning in OWL 2)

Current status of the reference ontology: PATO2YAMATO

•

Including about 1,000 PATO terms into YAMATO framework

•

Basic form of context-dependent ordinal values are defined. They are workable under the classification of organisms.

•

Basic form of quality representation (EAV and EQ) are already defined in YAMATO.

http://www.brc.riken.go.jp/lab/bpmp/ontology/ontology_pato2yato.html

Preliminary trial of simple conversion of EQ to EAV

1,450 EQ annotation: ( OBO cross-product file for Mammalian Phenotype ontology )

reference: PATO2YAMATO

EAV-quality representations in YAMATO framework

The ontology helps the automatic conversion from EQ to EAV!

We are planning full conversion of EQ across multiple species with coordinated EAV quality representation.

Summary of this talk

• •

This study shows: YAMATO’s framework helps to coordinate different “qualities” for phenotype information in both of reality and description level.

Role-model successfully coordinated ordinal values dependent on multiple contexts (deviation-based and simple comparison).

• • • •

Future views: Automatic conversion of EQ of multiple species to EAV.

Modeling of contexts of experimental conditions.

Integration of qualitative and quantitative phenotype data.

Coordination of more complicated phenotype data sets from multiple species and experiments.

Acknowledgements

RIKEN BioResource Center Nobuhiko Tanaka, Kazunori Waki, Terue Takatsuki University of Cambridge Georgios V. Gkoutos, Robert Hoehndorf NalaPro Technologies Inc Yoshihiro Okuda, Tatsuya Kushida Enegate corp Mamoru Ota RIKEN BASE Norio Kobayashi, Koji Doi, Tetsuro Toyoda Department of Knowledge Systems, ISIR, Osaka University Koji Kozaki, Riichiro mizoguchi

貴為和以 “

Harmony is to be valued.”

In “Seventeen-article constitution” (A.D 603, YAMATO imperial court in ancient Japan) Authored by Prince Sh ō toku (A.D. 573–621) Thank you !