II-SDV 2015, 20 - 21 April, in Nice

Dr. Matthias NegriScientific Information Center

Boehringer Ingelheim Pharma GmbH & Co. KG

Chemistry-Enriched Patent Curationsemi-automatic analysis and elaboration of patents

II-SDV Nice, 21 April 2015

Árpád FigyelmesiChemAxon

Content

1. Chemistry in patents

2. Why do we need a patent curation workflow?

3. Semi-automatic Patent Curation Workflow - Overview

4. Linked tools/technologies

5. ChemCurator (ChemCC)

6. Semi-automatic Patent Curation Workflow – Step by Step

7. Lessons learned, weak-points, limitations

8. Outlook

Negri Matthias, II-SDV 2015 2

Chemistry in patents

Chemistry appears within diverse form in patents:

1. TEXT - IUPAC names, common names, etc

2. IMAGES - embedded within or attached to the document

3. ATTACHMENTS (MOL/CDX)

4. TABLES

– as ONE-image file (tables with chemistry and bioactivity data)

– as chemistry-only image files embedded within table tags

5. Markush Structures/Formulas with R-groups

---------------------------------------------------------------------------------------

Currently NO commercial solution covers all these cases

Most of the cases are considered in the patent curation workflow

(Markush/R-group Formulas recognized and stored separately)


Why do we need a patent curation workflow?

Motivations:

1. Linked chemistry-retrieval from patents (+ chemistry as images)

2. IUPAC-enriched XML patent files as NEW source for text-mining

3. extraction of bioactivity data/targets/diseases/… in relation to chemistry

4. Similarity/Substructure frequency in compound sets of patents

5. …


Semi-automatic Patent Curation Workflow

Overview – current state

2 parallel branches


I2E API KNIME – Batch indexing, text-mining and (relational) data retrieval

SLOWER & memory intensive vs BUT Higher Quality, More Control & IUPAC-enriched XML

FASTER vs LESS informative/flexible - ChemCC as the (near) future perspectiveINPUT

Linked tools/technologies

1. KNIME/XPATH

2. ChemAxon ChemCurator (ChemCC)

3. Other ChemAxon tools in KNIME nodes (document2structure/d2s, Naming,

Molconverter, Structure checker, Standardizer, …)

4. Text/data-mining – Linguamatics I2E (+I2E Chemistry)

5. Optical Structure Recognition – Keymodule CLiDE Batch


Content








8. Outlook


Computer-aided chemical data extraction

English, Chinese and Japanese N2S

Markush Editor

Structure Checker

Hit visualization

Third party OSR technologies

ChemCurator (ChemCC)

Árpád Figyelmesi, II-SDV 20158


Name to Structure

Support for many nomenclatures (common, drug names, …)

IUPAC names

Custom dictionaries

English (2008)

Chinese (2013)

Japanese (2014)


Compound Extraction View

Compound listProject explorer

Annotated document

Selected structures


10

Markush Extraction View

Markush editor

Example structures

Annotated document

Project explorer

Selected structures

Structure checker


11

General Document Curation

Extract Markush Structures from patents

Extract specific structures

Journal articles

Company reports

Patent examples

Structure extraction wizards

Exclude fragments, chemical elements, etc.




Integration & Information Sharing

Other ChemAxon products:

Direct IJC schema connection

Project sharing function

Accessible from Plexus, IJC, etc.

Third party tools:

Standard file formats

Export functions

Easily processable projects


Content








8. Outlook


Semi-automatic Patent Curation Workflowa) input sources and b) bibliographic data

a) Input sources

files with patent-IDs list

XML collection

…

b) Retrieval of bibliographic information and attachment data

family ID, patent references, expiration date, etc

Attachment files MOL/CDX (US-patents only), TIF files

….


Semi-automatic Patent Curation Workflowc) chemistry retrieval/extraction/filtering

1. ChemCurator branch

data retrieval (XML, attachments) from IFI Claims Direct BI-server

ChemCurator project creation/sharing/annotation html output

Chemistry extraction name2structure/document2structure sdf output

Generation of pre-annotated patent set stored as ChemCC projects

Faster, but lower quality within the chemistry extraction process


2. KNIME branch

- OCR-errors CLEAN-UP in KNIME improved chemistry recognition

- MOL/CDX/TIF - standardizer, structure checker filter formulas, solvents, R-groups

Higher quality and more control in chemistry extraction process



2. KNIME branch

MOL IUPAC

CDX IUPAC

TIFF (via CLiDE) IUPAC



Merging and Comparison of the converted chemistry output of MOL/CDX/TIF – 2 “quality” checks

IUPAC

string length (different output order of chemicals in multiple molecules image/multiMOL files

OCR-correction (“dictionary” based)

2. KNIME - Chemistry “Normalization”

(within KNIME) set up a relation between each TIFF/attachment file

1. to (one or more) IUPAC name(s)

2. to a position/section in the text/document



Merge IUPAC Clean-Up IUPAC

If NO IUPAC IMG-name is set

“Normalize” IUPAC names

Semi-automatic Patent Curation Workflow d) TIF/attachment replacement with IUPAC names

Chemistry present as text is recognized and extracted either via

- Textmining (I2E chemistry – d2s is working in behind) or

- Within KNIME/ChemCC using annotate/molconvert

Replacement:<chemistry> vs IUPAC

IUPAC-enriched XML


OCR-errors in chemical names

Semi-automatic Patent Curation Workflow d) TIF/attachment replacement with IUPAC names

TIF

CDX

MOL

Replacement with the derived IUPAC name


XPATH/XML parsing and extraction of:

Tables

Rows - XML tags & strings

Entries - XML tags & strings

Semi-automatic Patent Curation Workflow e) Bioactivity/tabular data extraction with KNIME/XPATH


IUPAC-enriched XML as source for I2E API/textmining

indexing

pre-defined queries

results retrieval

saved as SDF files (KNIME)

Semi-automatic Patent Curation Workflow f) Text-/datamining with Linguamatics I2E via KNIME

Text-mining retrieved (chemistry-related) information

Example Nr.

Bioactivity data from tables

Claims, regions where chemistry appears in patents

Genes, diseases


1. Example Nr. – IUPAC

Table:Image:

For comparison – chemistry in PDF:

Semi-automatic Patent Curation Workflow f) Bioactivity Data using I2E multi-queries – 2 steps

Source: (IUPAC-enriched) XML

2. Example Nr. – Bioactivity data

24

IUPAC

Bioactivity

Example Nr.

Semi-automatic Patent Curation Workflowg) Visualize data-/textmining results in ChemCC

SDF file loaded into ChemCC project + automatic mapping to existing chemistry


Lessons learned, weak-points, limitations

1. Advantages KNIME Full-Mode (MOL/CDX/TIF) vs ChemCC branch

chemistry check/normalization – 3 input sources improved quality

improved chemistry recall - ALL images (incl. tables and drawings)

More filtering options in KNIME workflow vs ChemCurator only

IUPAC-enriched XML as new source for I2E

….


Lessons learned, weak-points, limitations

2. No full automation of the workflow due to lack of homogenicity in patent data (US vs WO, EP, etc..)

Missing attachment files

No tables present in XML

Error rate in chemistry recognition (OPSIN vs n2s/d2s)

…

NEEDS: different workflows/branches, patent-files clean-up (OCR)

3. Time & Computational Resources-consuming process


Outlook

1. KNIME Workflow

Add new data fields to Chemicals: BI-internal codes, genes, targets, etc..

Usage of ChemCC html output as source for textmining

Ontology mapping

Expand workflow by including other sources (internal PDF, literature full-text)

Use KNIME to interconnect to BI-intern workflows, DB, etc

chemistry-linked information in a patent-DB improved (semantic) search


Outlook

2. ChemCurator

Improved n2s

New command-line functions

Complex-phrase requests from IFI server

Improved SDF import

Preprocessing wizards


Thank You !


INPU

T

II-SDV 2015, 20 - 21 April, in Nice

Internet

sdv nice

outlooknegri matthias

patents chemistry

text iupac names

linked chemistryretrieval

common names

chemcurator chemcc6

drug names