BCB 444544

1
BCB 444/544
Lecture 16 Protein Function Prediction 16 Oct 1

• Thanks to Drena Dobbs for many borrowed
  modified PPTs

2
Required Reading (before lecture)

• Wed Oct 1 for Lecture 16
• Chp 7
• Fri Oct 3 for Lecture 17
• Chp 8

3
Homework Assignments

• HW3 on HMMs is posted on the course webpage under
  homework
• Due Oct 6th by 5pm
• 544 Students (and any in 444 who want to do a
  project)
• Extra HW1 posted
• Due Friday Oct 3rd by 11AM

4
Chp 7 - Protein Motifs Domain Prediction

• SECTION II SEQUENCE ALIGNMENT
• Xiong Chp 7
• Protein Motifs and Domain Prediction
• Motif Discovery in Unaligned Sequences
• Identification of Motifs Domains in MSAs
• Motif Domain Databases Using Regular
  Expressions
• Motif Domain Databases Using Statistical Models
• Protein Family Databases
• vSequence Logos

5
Protein Function

• Proteins are primary molecules responsible
• for carrying out cellular functions
• Most enzymes that catalyze chemical reactions are
  proteins (but some are RNAs!)
• Proteins have complex structures that are
  critical for their functions

Protein structure for dystrophin encoded by the
largest known gene in humans
6
Protein Structure 4 levels of organization
7
Key Aspects of Protein Function Localization
Interactions
Protein localization - function depends on
proteins being in right place at right time!
Protein interactions - function depends on
proteins interacting with correct partners inside
cells!
8
Protein Sequence-Structure-Function

• Amino acid sequence determines protein structure
• But some proteins need help folding
  ("chaperones") in vivo
• Proteins fold to a single "native" structure
  (under a specific set of conditions)
• Protein structure determines function
• But level, timing location of expression are
  important
• Interactions with other proteins, DNA, RNA,
  small ligands are also very important!!
• PROBLEMS
• We don't know the "folding code" that determines
  how proteins fold!
• We don't know the "recognition code" that
  determines how proteins find and bind their
  correct partners!

9
Motifs Domains

• Motif - short conserved sequence pattern
• Associated with distinct function in protein or
  DNA
• Avg 10 residues (usually 6-20 residues)
• e.g., zinc finger motif - in protein
• e.g., TATA box - in DNA
• Domain - "longer" conserved sequence pattern,
  defined as a independent functional and/or
  structural unit
• Avg 100 residues (range from 40-700 in
  proteins)
• e.g., kinase domain or transmembrane domain - in
  protein
• Domains may (or may not) include motifs

10
2 Approaches for Representing "Consensus"
Information in Motifs Domains

• Regular expression
• Statistical model

11
Regular Expressions

• Reduce information from MSA
• e.g., protein phosphorylation site motif
  S,T- X- R,K
• Symbols represent specific or unspecified
  residues, spaces, etc.
• 2 mechanisms for matching
• Exact
• "Fuzzy" (inexact, approximate) - flexible, more
  permissive to detect "near matches"

12
Statistical Models

• Includes probability information derived from MSA
• PSSM
• Profile
• HMM

13
Motif Discovery in Unaligned Sequences

• Expectation Maximization
• Gibbs Sampling

14
Expectation Maximization

• Generate "random" alignment of all sequences,
  derive PSSM, iteratively match individual
  sequences to PSSM to edit improve it
• Problems? Can hit a local optimum (premature
  convergence)
• Sensitive to initial alignment
• MEME - Multiple EM for Motif Elicitation -
  modified EM, avoids local optimum issues

15
Gibbs Sampling

• Generate "trial" PSSM from random alignment
  first, as in EM, but leave one sequence out of
  initial alignment, then iteratively match PSSM to
  left-out sequences
• Gibbs Sampler - web-based motif search via Gibbs
  sampling

16
Practical Advice
17
(No Transcript)
18
Motif Domain Databases

• Based on regular expressions
• Prosite (Interpro)
• Emotif
• Limitation these don't take probability info
  into account
• Based on statistical models
• PRINTS
• BLOCKS
• ProDom
• Pfam
• SMART
• CDART
• Reverse PsiBLAST

• READ your textbook try some of these at home
  there are distinct advantages/disadvantages
  associated with each
• TAKE HOME LESSON
• Always try several methods!
• (not just one!)

19
Prosite

• Database of protein families and domains
• Attempts to derive a signature for each protein
  family or domain that distinguishes the protein
  from all others
• Currently contains over 1000 signatures
• Uses both patterns (regular expressions) and
  profiles

20
Prosite Patterns

• Start with a MSA of related sequences pay
  special attention to things like active sites,
  residues that bind a ligand
• Try to find a conserved 4-5 amino acid sequence
  that includes these functional sites
• Scan Swiss-Prot DB with this sequence
• If Swiss-Prot hits are specific for the family,
  stop
• If not specific, lengthen the pattern, and search
  again

21
Prosite Profiles

• Prosite Patterns attempt to find a very short
  signature
• Prosite Profiles attempt to characterize the
  family or domain over the entire length
• Issue profile describes both conserved and
  non-conserved regions so it is possible to match
  the profile but not be a member of the family
• Prosite tests its profiles to try to avoid this
  problem

22
Emotif

• More than 170,000 highly specific and sensitive
  protein sequence motifs
• Uses regular expressions
• Based on MSA of related proteins
• Enumerates all possible motifs, not just one per
  family or domain
• Sensitivity is defined by how many of the
  sequences in the MSA fit the motif
• Specificity is based on how often you would
  expect to see the motif in randomly generated
  sequences

23
PRINTS

• Collection of fingerprints a set of motifs
  that can predict the occurrence of other motifs
• Uses a MSA with only a small number of sequences
  and attempts to define fingerprints
• These fingerprints are used to scan their
  database for more sequences containing the
  fingerprint
• Rebuild the fingerprints, and iterate until the
  fingerprint converges

24
Protein Family Databases

• In addition to databases of "related" protein
  sequences, based on shared motifs or domains
  (Pfam, BLOCKS, CDART), some databases "cluster"
  sequences into families based on near full-length
  sequence comparisons
• COGs - Clusters of Orthologous Groups (at NCBI)
• Mostly Prokaryotic sequences
• KOG newer Eukaryotic version
• COGnitor - software to search database
• ProtoNet - also clusters of homologous protein
  sequences
• Advantages tree-like hierarchical structure
• Provide GO (gene ontology) annotations
• Provides InterPro keywords

25
Pfam

• Contains over 10,000 protein families
• Pfam-A high quality, manually curated
• Pfam-B automatically generated, comprehensive
  collection
• Uses 2 profile HMMs to describe each protein
  family
• Why 2?
• One for global matches, one for local matches

26
CDART

• Conserved Domain Architecture Retrieval Tool
• Given a query sequence
• Identifies functional domains
• Lists proteins with a similar domain architecture
  sequential order of conserved domains
• Can detect remote homologs domains are
  conserved, but sequence of each domain may vary
  quite a bit

27
PIRSF

• Protein Information Resource SuperFamily
  classification system
• Network classification system based on
  evolutionary relationships of whole proteins
• Advantages
• Allows for both generic and specific functions to
  be assigned
• Allows for classification of proteins without
  well-defined domains

28
PIRSF

• Multiple levels of classification
• Superfamily
• one or more common domains
• Homoemorphic family
• proteins that are homologous full-length
  sequence similarity AND common domain
  architecture
• Homeomorphic subfamily
• functional specialization

29
What we didnt talk about

• Machine learning!
• Given a protein sequence, predict the function
• Many, many methods exist, no clear favorite
• Most are specific for some type of function,
  e.g., kinase
• In lab Jafa metaserver sends your sequence to
  5 different prediction servers and combines the
  results into a single prediction

30
What we didnt talk about

• Protein function prediction based on protein
  structure
• Coming later