Chemoinformatics Theory

1
Chemoinformatics Theory
Yoon Soo Pyon ysp2_at_case.edu October 19th, 2007
2
Outline

• Chemoinformatics-What is it?
• Molecular descriptors and chemical spaces
• Chemical spaces and molecular similarity
• Molecular similarity, dissimilarity, diversity
• Modification and Simplification of chemical
  spaces
• Compound Classification and Selection
• Similarity Searching
• Machine Learning Methods
• Library Design
• Quantitative Structure Activity Relationship
  Analysis (QSAR)
• Virtual Screening and compound filtering

3
Chemoinformatics-What is it?

• Use of computer and informational techniques,
  applied to a range of problems in the field of
  chemistry.
• This in silico techniques are used in
  pharmaceutical companies in the process of drug
  discovery.

4
Chemoinformatics-What is it?
5
Chemoinformatics-What is it?
6
Molecular descriptors and chemical spaces

• Chemical reference spaces where molecular data
  sets are projected and analysis of design is
  carried out.
• Definition of chemical spaces critically depend
  on the use of computational descriptors of
  molecular structure, physical or chemical
  properties.

7
Molecular descriptors and chemical spaces
8
Molecular descriptors and chemical spaces

• There are no generally preferred descriptor
  spaces.
• Require to generate reference spaces for
  specific application on a case by case

9
Chemical spaces and molecular similarity

• Similar Property Principle Molecules having
  similar structures and properties should also
  exhibit similar activity. (Often but not always
  true)
• Thus, molecules that are located closely
  together in chemical reference space are often
  considered to be functionally related.

10
Chemical spaces and molecular similarity
11
Molecular similarity, dissimilarity, and diversity

• Diversity analysis
• Select different compounds from a given
  population
• Evenly populate a given chemical space with
  candidate molecules. Only selecting compounds
  that are at least a pre-defined minimum distance
  away from others.
• Dissimilarity Inverse of molecular similarity
• Dissimilarity analysis played a major role in the
  pharmaceutical industry.

12
Molecular similarity, dissimilarity, and diversity

• Dissimilarity algorithm
• Select a subset of k maximally dissimilar
  compounds
• ? due to combinatorial problem, non-trivial
  challenge
• Other dissimilarity algorithm
• Decide on a desired size, n, of a final subset
• Select a seed compound and place it in the
  subset
• Calculate the dissimilarity between each of the
  other compounds and those in the subset
• Choose the next compound as the one most
  dissimilar to those in the subset
• If fewer than n in the subset, repeat the
  calculation of the dissimilarity until n is
  achieved
• Complexity varies as the square of n

13
Modification and Simplification of Chemical
Spaces

• High dimensional chemistry space might often too
  complex for carrying meaningful analyses.
• Why?
• 1) Major areas of high dimensional chemical space
  might not populated and remained as empty.
• 2) Correlation effects between selected
  descriptors dramatically distort the reference
  space.
• Therefore,
• 1) Design low-dimensional reference spaces
• 2) Simplify high-dimensional spaces
• 3) Reduce their dimensionality

14
Modification and Simplification of Chemical
Spaces (contd.)

• Auto scaling or variance scaling
• Why? Descriptor with large value range will
  dominate those having smaller one.
• Dimension reduction

15
Modification and Simplification of Chemical
Spaces (contd.) Dimension reduction

• Assumption High dimensional descriptor spaces
  have at least some intrinsic redundancy.
• Two approaches
• To identify those descriptors that are most
  important for representing the original dataset
  and the relationships they form between objects
  for lower-dimensional representation
• ex) multi dimensional scaling (Agrafiotis,
  et al. 2001)
• To attempt to generate new descriptors for
  lower-dimensional spaces by combining important
  contributors from original one.
• ex) Principal Component Analysis (PCA)

16
Modification and Simplification of Chemical
Spaces (contd.) - Simplification

• Simplification of n-dimensional descriptor
  spaces
• ex) Binary descriptor transformation
• above mean ? 1, below mean ? 0

17
Compound Classification and Selection- CLUSTER
ANALYSIS

• Aim is to divide a group into clusters where
  objects in the cluster are similar, but objects
  in other clusters are dissimilar
• Many algorithms for doing this
• Hierarchical methods seem to be better than
  non-hierarchical
• Sometimes called a distance-based approach to
  compound selection, because distance is measured
  between pairs of compounds

18
Compound Classification and Selection- CLUSTER
ANALYSIS
19
Compound Classification and Selection-
Hierarchical Clustering

• The composition of each cluster depends on the
  one from which it was derived
• Agglomerative methods start at the bottom and
  merge similar clusters (bottom-up)
• Wards method clusters are formed to minimize
  the variance (i.e., the sum of the squared
  deviations from the mean)
• Others centroid method and the median method
• Divisive hierarchical clustering starts with all
  compounds in a single cluster and partitions the
  data (top-down)

20
Compound Classification and Selection-
Non-Hierarchical Clustering

• Organize compounds into an initially defined
  number of independent clusters.
• Methods
• nearest neighbor Jarvis Patrick clustering
• relocation K-means

21
Compound Classification and Selection-
Partitioning

• Rather than comparing molecular positions,
  establish a coordinate ore reference system in
  chemical space.
• Compounds that populate the same partitions
  considered to be similar.

22
Compound Classification and Selection-
Partitioning

• Diversity-based selection - Aims at generating a
  small representative subset of a compound
  collection. It is attempted to generate evenly
  populated partition.
• Activity-based selection Known active compounds
  are added to the source database prior to
  partitioning. Compounds in database mapping close
  to known activities are then selected as
  candidate for testing to identify new hits.

23
Compound Classification and Selection-
Statistical Partitioning

• Recursive partitioning most popular statistical
  partitioning. A decision tree method
• Divides datasets along decision trees formed by
  sequences of molecular descriptors.
• ex) The compounds could be divided according to
  molecular weight.

24
Compound Classification and Selection-
Statistical Partitioning

• Statistical partitioning methods such as
  recursive partitioning is also very attractive
  tools for the analysis of HTS data sets.

25
Similarity Searching Structural queries and
graphs

• Detection of structural fragments or
  substructures is a simple but popular form of
  similarity searching.

26
Similarity Searching Structural queries and
graphs

• Contemporary substructure search methods are
  mostly based on dictionaries of predefined
  molecular fragments.
• Queries can be transformed into an
  machine-readable format such as Simplified
  Molecular Input Line Entry Specification (SMILES)
  code.
• SMILES encodes 2D representation of molecules as
  linear strings of alpha-numeric characters.

27
Similarity Searching Structural queries and
graphs (SMILES)
28
Similarity Searching Structural queries and
graphs

• Subgraph-isomorphism
• Common substructures can also determined by
  systematic mapping of corresponding node
  positions in graph.
• However, computationally expensive
• Reduced graph
• Nodes do not represent atoms but features such as
  functionally important groups or whole ring
  system.
• Become more suitable for node matching procedures
  and similarity searching.

29
Similarity Searching Structural queries and
graphs (Reduced graph )
30
Similarity Searching Pharmacophore

• A molecular framework that carries the essential
  features responsible for drugs biological
  activity
• Spatial arrangements of atoms or groups that are
  responsible for biological activity
• Often used as 3D queries for database searching

31
Similarity Searching Fingerprints

• Fingerprints
• widely used similarity search tools.
• consist of various descriptors that are encoded
  as bit strings
• Bit strings of query and database compared using
  similarity metric such as Tanimoto coefficient

32
Machine Learning Methods

• Important role in chemoinformatics
• For example, it is usually difficult to predict
  which types of descriptors are most suitable for
  a given search, classification.
• Therefore, machine learning techniques are often
  used to facilitate descriptor selection
• Applied to generate complex predictive models by
  iterative processing of molecular learning sets
• Genetic algorithms
• Neural Networks
• Self Organizing Maps (SOM)

33
Machine Learning Methods Genetic algorithms

• Different parameters and model solutions to given
  problems are encoded in a chromosome and
  subjected to iterative random variation, thus
  generating a population.
• Solutions provided by these chromosomes are
  evaluated by fitness function that assign high
  scores to desired results.
• Chromosomes yielding best intermediate solutions
  are subjected to mutation and crossover operation
  that correspond to random genetic mutations and
  gene recombination events.
• The resulting modified chromosomes represent the
  next generation and the process is continued
  until the obtained results meet a satisfactory
  convergence criterion

34
Library Design

• Diverse Library
• Focused Library

35
Quantitative Structure Activity Relationship
Analysis (QSAR)

• Goal Evaluation of molecular features that
  determine biological activity and the prediction
  of compound potency as a function of structural
  modification

36
Virtual Screening and Compound Filtering

• VS(Virtual Screening) - the process of screening
  large databases on the computer for molecules
  having desired properties and biological
  activity.
• A major application of VS techniques is the
  identification of novel active molecules in large
  compound databases.
• Series of known active compounds are added as
  search templates to a source DB and then
  compounds that are identified as similar to these
  templates based on VS calculations are selected
  as candidate molecules for experimental
  evaluation

37
(No Transcript)
38
Virtual Screening and Compound Filtering- Filter
Functions

• Filter functions are very popular tools for VS
• Attempts to identify compounds with desired
  properties and discard others.
• Have been implemented for analysis of diverse
  molecular properties including chemical
  reactivity, toxicity, drug-like character,
  absorption, distribution, metabolism, excretion
  (ADME) parameters.
• Ex) Aqueous solubility, Passive absorption
• blood-brain-barrier penetration,
  metabolic stability,
• oral availability

39
Virtual Screening and Compound Filtering- Filter
Functions
40
Thank You