milkER a milk informatics resource

1
milkER a milk informatics resource

• Stephen Edwards BSc.
• University of Edinburgh
• BioNLP meeting 6th June 2005

2
Overview

• Aims of milkER
• milkER database
• Text-mining
• Potential targets

3
milkER aims

• To amalgamate disperse milk information into one
  resource, allowing more focused analysis of milk
  proteins in relation to dairy issues, health and
  disease.

4
A milk database

• Knowledge on milk affects many industries
• UniProt, GenBank excellent resources
• Marsupial genomics database (New Zealand)
• Glasgow genomics data
• Chinese database
• Polish bioactive peptide database
• Food property database (commercial)

5
Milk components

• Fat, carbohydrates, proteins, minerals
• Growth factors, enzymes, enzyme inhibitors,
  immunoglobulins, allergens, disease factors,
  anti-bacterial proteins, opioids
• 1. Deliberate
• 2. Leakage from blood
• 3. Result of disease conditions
• 4. Engineered
• 5. Bacterial origin

6
milkER database

• Database using BioSQL which allows incorporation
  of UniProt, EMBL, GenBank entries

7
LOCUS NM_173929 790 bp
mRNA linear MAM 27-OCT-2004 DEFINITION Bos
taurus lactoglobulin, beta (LGB), mRNA. ACCESSION
NM_173929 VERSION NM_173929.2
GI31343239 KEYWORDS . SOURCE Bos taurus
(cow) ORGANISM Bos taurus
Eukaryota Metazoa Chordata Craniata
Vertebrata Euteleostomi Mammalia
Eutheria Cetartiodactyla Ruminantia Pecora
Bovidae Bovinae Bos. REFERENCE 1
(bases 1 to 790) AUTHORS Jayat,D.,
Gaudin,J.C., Chobert,J.M., Burova,T.V., Holt,C.,
McNae,I., Sawyer,L. and Haertle,T.
TITLE A recombinant C121S mutant of bovine
beta-lactoglobulin is more
susceptible to peptic digestion and to
denaturation by reducing agents and
heating JOURNAL Biochemistry 43 (20),
6312-6321 (2004) PUBMED 15147215 REMARK
GeneRIF Results suggest that the stability of
beta-lactoglobulin arising from the
hydrophobic effect is reduced by the C121S
mutation so that unfolded or partially
unfolded states are more
favored. ORIGIN 1 actccactcc
ctgcagagct cagaagcgtg atcccggctg cagccatgaa
gtgcctcctg 61 cttgccctgg ccctcacctg
tggcgcccag gccctcatcg tcacccagac catgaagggc
..
8

Information retrieval
Other Databases
EMBL
UniProt
Information extraction
Other Sources (e.g. published tables)
milkER population
milkER
Web Query
9
milkER database

• Database using BioSQL which allows incorporation
  of UniProt, EMBL, GenBank entries
• Library of literature on milk
• User interface (www.milker.org.uk)

10
(No Transcript)
11
(No Transcript)
12
(No Transcript)
13
Text-mining

• Machine reading of text
• Many techniques involved
• Tokenisation
• Stemming (Activation ? Activat)
• POS tagging (Protein ? noun)
• Abbreviation expansion (CN ? Casein)
• Entity identification (Casein ? protein)
• Dictionary

14
Increased levels of IgA antibodies to B-LG were
found and were shown to be an independent risk
marker for type 1 diabetes.
Increased past participle levels
plural noun of preposition
Tokeniser / POS tagger
IgA antibody B-LG protein Diabetes disease
Entity identification
Parser
IgA antibodies to B-LG MARKER type 1
diabetes
15
Information extraction

• Rule based
• interact bind activate
• protein (0-5 words) verbs (0-5 words)
  protein
• (Blaschke and Valencia, 2002)
• Machine-learning
• Statistical methods, Hidden Markov Models
• Learn interfillers, text lying between tagged
  entities (Bunescu et al, 2004)

16
Difficulties

• Synonyms
• Proteins and genes with same name
• Funny names e.g. ERK-1/2, and gene!
• Variability of natural language
• Compounded names
• Co-ordination, negatives, speeling errors

17
Evaluation

• Precision (P) - how correct is output
• Recall (R) - how often does it pick
• F-measure - combines P and R
• IE systems can achieve high results, but not
  enough to populate databases automatically

18
Text-mining uses

• Aim to extract interactions and diseases
• Swanson (Fish oil)
• Srinivasan (Turmeric)

19
General model for discovering implicit links
between topics Starting topic Turmeric
(inhibits) Intermediate topic Nuclear
factor-kappa B (involved in) Terminal
topic Crohns disease
Diagram taken from Srinivasan et al, 2004
20
Targets for text mining

• Many milk relationships still require further
  investigation
• Positive reasons
• - nutritional benefits
• - neonatal growth
• - antimicrobial activity
• - bioactive peptides

21
Targets for text mining (cont.)

• Negative reasons
• - recent link with Alzheimer's
• - diabetes link
• - asthma
• - human reactions to cow hormones
• (e.g. Acne, Danby 2005)
• - drug transfer to milk and effects
• - allergic reactions/intolerance
• - toxic contaminants

22
milkER process

• 897 proteins, 772 dna, 1232 rna
• Analyze references (1465 MEDLINE refs)
• MeSH terms, GO terms etc
• POS tag
• UMLS standardisation
• Gene/protein dictionary
• Extract relations

23
Milk literature
24
milkER interactions

• Table of interacting proteins
• Store as queryable XML strings?
• Discover links between proteins and disease
• Create hypotheses
• Confirm experimentally

25
Diabetes

• Pancreas secretes hormones
• Glycagon, increases conversion glycagon ? glucose
• Insulin, increases conversion glucose ? glycagon.
  Allows glucose into cells.
• Condition where the amount of glucose in the
  blood is abnormally high as the body cannot use
  it adequately as fuel

26
Diabetes

• Affects 3-5 of industrialised populations
• Type 1 (10)
• Genetic and environmental factors (e.g. diet)
• Decreased insulin production
• Mostly develops lt age 20
• Type II (90)
• Resistance of body to insulin
• Normally develops gt age 40
• Often associates with high B.P, cholsterol and
  arterial disease

27
Milk and diabetes
28
Selected quotes

• More research is needed on all aspects of
  lactation in women with diabetes.
• Reader D. et al, Curr Diab Rep. 2004
• The effect of high protein intakes from
  different sources on glucose-insulin metabolism
  needs further study
• Hoppe et al, European Journal of Clinical
  Nutrition 2005
• American children also tend to be heavier than
  those from European countries, skewing the
  growth charts further.
• The Scotsman Sat 5 Feb 2005
• The government currently recommends that babies
  should be fed breast milk alone for the first six
  months - the WHO recommends two years.

29
Conclusions

• Knowledge of milk vital in many areas
• milkER aims to bring disparate milk data together
• Text-mining can wade through large amounts of
  data to retrieve and discover vital information

30
Future work

• Relation extraction of milk literature
• Extend content of milkER to include interaction
  data
• Create hypotheses for experimental work

31
Acknowledgements

• Prof. Lindsay Sawyer
• Dr. Carl Holt (Hannah Research Institute, Ayr)
• Prof. Bonnie Webber (Informatics)
• Dr. Alistair Kerr and Dr. Douglas Armstrong for
  technical support

32
References

• Acne/milk
• Acne and milk, the diet myth, and beyond (Danby,
  2005)
• Diabetes/milk
• Milk and diabetes (Schrezenmeir et al, 2000)
  REVIEW
• The role of ?-casein variants in the induction of
  insulin-dependent diabetes (Elliott et al, 1997)
• Text-mining
• Natural language processing and systems biology
  (Cohen et al, 2004) REVIEW
• Mining MEDLINE for implicit links between dietary
  substances and diseases (Srinivasan et al, 2004)
• Learning to extract proteins and their
  interactions from MEDLINE abstracts (Bunescu et
  al, 2003)