Download slides - ChemAxon

IBM Research 1 © 2011 International Business Machines Corporation 1

IBM Research

Source – J Kreulen


IBM Research

Computer Curation of Patents &

Scientific Literature - (Information Analytical Services )

Transforming Information Into Value

Stephen K Boyer Ph.D.

Collabra, Inc [email protected]

[email protected]

408-858-5544

mailto:[email protected]

mailto:[email protected]


IBM Research

Patents also have (Manually Created) Chemical Complex Work Units (CWU’s)

As text

Chemical names found in the text of

documents

As bitmap images

Pictures of chemicals found in the document

Images

Patents (and scientific papers) contain molecular data in many different forms -


IBM Research

4

– Computer Curation & Analysis

Alert

Understand

Discovery

Point of View

Analyzing Taxonomies Analyzing Relationships

Create Point of View

Persistence Queries

Related Databases

Analyzing Sentiment Analyzing Influencers

Discovery

Monitor

Discovery

Topics

Visual / Faceted Search

Create the Landscape

Backend - Building the System Frontend - Discovery

Dat

a So

urc

es

U.S. Patents (1976 -—

2009)

U.S. Pre-

Grants (All)

PCT & EPO Apps

Medline Abstracts

(>18 M)


IBM Research

Computer curation now involves multiple types of analysis

• Analysis of text

• Analysis of image

• Analysis of XML files

Derived Meta data

Internal data

IBM + Collaborator input

Output db to Collaborators

• Analysis of (CWU’) s


IBM Research

Chemical from Complex Work Units

Chemicals from Image Analysis

Chemicals from Text Analysis

Chemical data derived from multiple computer curation processes

•text analysis •image analysis •chemical complex work units images -

Due to the nature of the data & the limitations of the technology –

Post processing workflows are required to clean up the raw chemical

data derived from our 3 processing streams


IBM Research

Chemical from Complex Work Units

Chemicals from Image Analysis

Chemicals from Text Analysis

Chemical data derived from multiple computer curation processes

•text analysis •image analysis •chemical complex work units images -

Due to the nature of the data & the limitations of the technology –

Post processing workflows are required to clean up the raw chemical

data derived from our 3 processing streams

We now use ChemAxon tools (libraries ) –to clean-up the chemical data,

prepare it for indexing – and for chemical search – (using DB2) .


IBM Research

Dat

a So

urc

es

View selected

Documents & Reports

U.S. Patents (1976 -—

2009)

U.S. Pre-

Grants (All)

PCT & EPO Apps

Medline Abstracts

(>18 M)

Selected Internet Content

User Applications

In-House

Content

Knime or Pipeline Pilot

BIW

SIMPLE

Chem Axon Search

Cognos/DDQB/ Other Apps

Parse & Extract

data

Annotator 1

Annotator 2

Database

+ compu ted Meta Data

e Classifier & Other Data Associations

Annotation Factory

Computational Analytics

ChemVerse (Semantic

Associations)

Computer Curation Process Overview

IP Database (e.g. DB2)

ADU*

* ADU = Automated Data Update

ChemVerse

db

ChemVerse

Services Hosted at IBM Almaden


IBM Research

Johnson & Johnson

Novartis

Bristol Myers

Squibb

Merck

By comparing the most relevant

concepts in patent data, we observed

patterns emerging.

Genentech is staking out white space

in the areas not covered by the other

major pharmaceuticals.

Pharmaceutical Industry Patent Landscape

AstraZeneca

Amgen

Total # of

patents

Pfizer

Merck

BMS

Novartis

J&J

AstraZeneca

Amgen

Total # of

patents

Pfizer

Merck

BMS

Novartis

J&J

Looking at US patent data for the last 18

years shows how pharmaceutical

companies are positioning themselves in

the market.

Leading companies like Pfizer,

AstraZeneca, and Amgen are increasing

their patent activity while other companies

are decreasing.


IBM Research Search Chemical Search using ChemAxon w/ DB2

Proximal Search Nearest Neighbor Search


IBM Research

BioTerm Analysis

Clustering Claims Originality

Discovery


IBM Research

Landscape Analysis

Visualization

Networks


IBM Research

Looking Forward …..From Simple -------to--------- DeepQA

Moving SIMPLE – to –SIIP (IBM- Global Businesses Services )

Using our base to move into DeepQA

Aggregating the data for Cognitive Computing


IBM Research

DeepQA: Massively Parallel Probabilistic Architecture

Question/T

opic

Analysis

Question

Hypothesis & Evidence

Scoring

Answer,

Confidence

Synthesis Final Merging

& Ranking

Query

Decomposition

Hypothesis

Generation

Hypothesis &

Evidence Scoring

Soft

Filtering

Hypothesis

Generation

Hypothesis &

Evidence Scoring Soft

Filtering

Hypothesis

Generation

. . .

Trained

Models

Primary

Search

Candidate

Answer

Generation

A.

Sources Supporting

Evidence

Retrieval

Deep

Evidence

Scoring

Answer

Scoring

E.

Sources

Evidence

Retrieval

Deep

Evidence

Scoring

14

DeepQA generates and scores many hypotheses using an extensible collection of Natural Language Processing,

Machine Learning and Reasoning Algorithms. These gather and weigh evidence over both unstructured and

structured content to determine the answer with the best confidence.



IBM Research

Factors to

consider

Current Watson capability

tuned to play Jeopardy

What is needed to purpose configure

Watson for specific Application

Knowledge Domain:

Open vs. Closed

Open domain to consider general

knowledge

Need to have domain-specific ontologies for specific

applications

Availability of

Knowledge Sources

Extensive availability of knowledge

sources

Develop/assemble sufficient knowledge sources for the

application at hand.

Training Data Watson used 20 years worth of

Jeopardy data to train the system

Availability of sufficient sample data for training is

critical to the performance of Watson.

Speech to Text Watson can handle some speech-to-

text but it is not perfect

It may be better to go first with just text and then

integrate speech-to-text capabilities

Response Time

(Latency)

Watson used 15TB of system memory

and massive processing to respond

within 3 seconds for Jeopardy

What is the application requirement for response time?

What accuracy is required? This will drive how Watson

will be configured.

Language support

(Cross/Multi Lingual)

Watson’s core algorithms are

language agnostic. The jeopardy

application supports only English.

Language-specific parsers will have to be developed if

Watson has to support multiple languages.

Question

Type

Watson processes by decomposing

and classifying the questions. Most of

the questions in Jeopardy are fact-

based questions.

What type of questions are relevant for the application?

If it requires special instructions to understand or using

audio/visual aids then they are not good candidates for

Watson.

Answer

Type

Watson is designed to be a powerful

general purpose natural language QA

system. It is not designed to be task-

oriented.

Watson can handle multi-step processing but it is not

suitable for process-oriented questions (e.g. did we

complete all preceding tasks A and B before starting

task C)

Quick view of considerations in applying Watson-type capabilities



IBM Research

InfoSphere Warehouse DB2, Informix, Netezza

Aggregating and storing data and content

InfoSphere BigInsights “Big Data” analysis (Hadoop)

IBM Content Analytics Natural Language Processing and content analysis leveraging UIMA

InfoSphere Streams Massively parallel analysis

IBM Power Systems Thousands of parallel processes

Related Innovations

Business Analytics BI, Predictive Analytics

and more

IBM Global Business Services

Research, expertise and analytical assets

ECM Solutions IBM eDiscovery Analyzer

IBM Classification Module IBM OmniFind Enterprise Search

Used by Watson

Workload Optimized Systems Integrated, Optimized by Workload

The Components of Deep QA



IBM Research

Sample Scenario from Literature

How do we create a drug that positively affects the "PI3K pathway“?

Ref: http://dipbsf.uninsubria.it/monti/BFPN%202009/nrd1902.pdf

An approach to seed the brainstorming: 1. Identify components of PI3K pathway, biological entities such as "cell reproduction"

and "growth regulation" 2. Identify related genes and proteins to these concepts using statistical affinity 3. Find chemicals known to affect these genes/proteins (using statistical associations) 4. Find drugs most similar to these chemicals (using drug dictionary and chemical

similarity)


http://dipbsf.uninsubria.it/monti/BFPN%202009/nrd1902.pdf


IBM Research

Cloud / Enterprise Integration

Hadoop++ (GPFS, sys mgmt, meta-tracker)

JSON Datamodel, JAQL

System T Entity extraction

Entity resolution

Unstructured

(item descriptions)

Semi-Structured

(clickstream)

Structured

(transaction logs)

Parallelized Algorithms

MSA Services

Enriched

Data Indices

Data Access Services

Analytical Services DeepQA

COGNOS

SPSS

iLOG

. . .

Solution Web Application Data and Analytics

Web Services

MetaTracker++ (Scheduling and Sys Admin)

Enterprise Applications Data and Analytics

Services




Entity resolution

Unstructured

(item descriptions)

Semi-Structured

(clickstream)

Structured

(transaction logs)


MSA Services Enriched

Data Indices



COGNOS

SPSS

iLOG

. . .

ISVs

MetaTracker++ (Scheduling and Sys Admin) Solution,

Analytics

and Data

Cloud

Enterprise



IBM Research

Cloud / Enterprise Integration




Entity resolution

Unstructured

(item descriptions)

Semi-Structured

(clickstream)

Structured

(transaction logs)


MSA Services

Enriched

Data Indices



COGNOS

SPSS

iLOG

. . .

Solution Web Application Data and Analytics

Web Services

MetaTracker++ (Scheduling and Sys Admin)

Enterprise Applications Data and Analytics

Services




Entity resolution

Unstructured

(item descriptions)

Semi-Structured

(clickstream)

Structured

(transaction logs)


MSA Services Enriched

Data Indices



COGNOS

SPSS

iLOG

. . .

ISVs

MetaTracker++ (Scheduling and Sys Admin) Solution,

Analytics

and Data

Cloud

Enterprise

Chem Axon

an IBM Vendor

for

Chemical

Cleanup,

Searching

& Analysis !


IBM Research

I would like to acknowledge the IBM Almaden Research – team

Jeff Kreulen

Ying Chen

Scott Spangler

Alfredo Alba

Tom Griffin

Eric Louie

Su Yan

Issic Cheng

Prasad Ramachandran

Bin He

Ana Lelescu

Qi He

Linda Kato

Ana Lelescu

Brad Wade

John Colino

Meenakshi Nagarajan

Timothy J Bethea

German Attanasio

+ a host of folks from

IBM China Labs -


IBM Research

Thank you.

IBM Almaden Research Center
