Challenges in Bioinformatics Part II

1
Challenges in BioinformaticsPart II

• Vasileios Hatzivassiloglou
• University of Texas at Dallas

2
More observations on GloFish

• GloFish license prohibits
• intentional breeding
• sale/barter/trade of offspring
• Attitudes on recombinant DNA / genetic
  engineering in Europe
• Le Monde Diplomatique (January 2004)
• Frankenfish and the future

3
Sequence analysis

• Given a genome sequence, determine the gene
  regions
• Uses data from observed output (expression)
• Models based on transducers and Markov Models

4
String Similarity

• Two issues
• Given two sequences, what is their similarity?
• How are two similar sequences aligned?
• Global alignment algorithms based on dynamic
  programming
• Local alignment FASTA, BLAST
• Useful for homology search (finding related genes
  / proteins in different organisms)

5
Motif finding

• Motifs are short sequences in DNA or proteins
  that are over-repeated in a sample
• Find motifs and understand their role
• Typically with statistical word counting
  methods
• Alternatively, with a predictive model and the EM
  (expectation-maximization) algorithm
• Important for
• recognizing promoter regions

6
Clustering

• Genes that are co-expressed in a DNA microarray
  experiment may be functionally related
• Cluster genes on
• sequence similarity, or
• expression data
• Issues include feature selection, outlier
  detection, validation of clusters (mathematically
  and biologically)

7
Classification

• Detect particular types of proteins or genes
  (e.g., find the gene responsible for a disease,
  find highly active proteins)
• HIV and leukemia as applications
• Multiple models based on different feature spaces
• Many algorithms including decision trees, kNN,
  and support vector machines (SVM)

8
Protein folding

• Proteins fold in a native state
• 3D properties important for drug interactions
  (pharmacogenomics)
• Variations from the native state can be
  indicators of disease (Alzheimers, mad cow
  disease)
• Folding depends on chemical interactions and
  thermodynamics
• Direct simulation impractical (1 day for 1ns)
• Treat as an optimization problem with approximate
  solutions (minimize an energy function)

9
Protein Folding Prior Knowledge

• Knowledge from other proteins can help
• Find similar primary structure in other proteins
  with known folding
• Folding can be predicted (at least locally) and
  experimentally verified

10
Phylogenetics

• Recovery of the tree of life
• Important for molecular biology because
• we can predict missing sequences from known
  sequences in related species
• we can predict function from related
  genes/proteins in another species

11
Phylogenetics

• Goal Classify species based on data
• Cladistics
• Characters are the differentiating features which
  have different states (e.g., flower color)
• Construct matrix of features and automatically
  locate best discriminating feature
• Misleading evidence can exist (homeoplasies) due
  to convergent evolution, e.g.,
• wings in insects and birds

12
Phylogenetics Other approaches

• Maximum likelihood models
• Estimate probability of change for each
  character, and arrange species in maximum
  likelihood tree
• Molecular systematics
• Measure similarity between short sequences of DNA
  (haplotypes) and use clustering to create the
  tree
• Can also use mRNA

13
Phylogeny of the dog (1997)

• Haplotype based on a sequence of 261 bp
• Compared 72 dog breeds and 27 geographically-based
  groups of wolves as well as a control group of
  other canids (coyotes, jackals)
• 27 haplotypes for dogs, 26 for wolves
• Maximum differences 10 (dog), 12 (wolf), 12
  (dogwolf)
• Minimum difference 20 (dog vs. other canids)
• Evidence shows
• Dog evolved from wolf
• Four classes of dogs, some more homogeneous
• Evidence conflicts with current dog breed
  classification

14
Other important computational issues

• Data storage
• Efficient database access and search
• Text search and information retrieval
• Lossless compression
• Database interactions
• Representation and visualization
• Image processing
• Robotics

15
Curation versus discovery

• Much easier (and faster) to have an expert check
  system results, than produce such results from
  scratch
• Automated discovery followed by curation
• increases thoroughness
• potentially removes bias (assuming system is not
  biased)

16
Curation

• Experimental results need to be verified by
  experts
• This is a large and time-consuming task
• How can it be facilitated?
• Interface issues, access to primary data, access
  to literature
• Concurrent verification
• Can it be modeled and automated?
• AI and statistical models

17
Knowledge modeling

• Models of biological processes and the steps in
  them (e.g., actions in regulatory networks)
• Needed to support automated processing of
  extracted data
• Distinct from data extraction

18
Examples of modeled knowledge

• Types of protein actions (bind, activate,
  phosporylate, ...)
• Constraints on actions
• (Functional) classes of proteins
• Ontologies of concepts in the biological domain
• Much of this is derived via text mining