Title: The Domain Structure of Proteins: Prediction and Organization.
1The Domain Structure of Proteins Prediction and
Organization.
- Golan Yona
- Dept. of Computer Science
- Cornell University
- (joint work with Niranjan Nagarajan)
Golan Yona, Cornell University
2PDB 1a8y 367aa long MKIIRIETSRIAVPLTKPFKTALRTVYTA
ESVIVRITYDSGAVGWGEAPPTLVITGDSM
3The domain structure of a protein
- A domain is considered the fundamental unit of
protein structure, folding, function, evolution
and design. - Compact
- Stable
- Folds independently?
- Has a specific function
4A protein is a combination of domains
Protein1 Protein2 Protein3
5Any signals that might indicate domain boundaries?
- A very weak signal if any in the sequence
- Usually domain delineation is done based on
structure - Best methods available manual!
- But structural information is sparse..
6Definitions and assumptions
- Domain continuous sequence that corresponds to
an elemental building block of protein folds. - A subsequence that is likely to be stable as an
independent folding unit. - Was formed as an independent unit, and later was
combined with others more complex functions. - There are traces of the autonomous units..
7First step..
- Gather data database search
- Histogram of matches is informative but noisy
- Mutations, insertions, deletions, conflicting
evidence
sequence
8Previous methods
- Methods based on the use of similarity searches
and knowledge of sequence termini to delineate
domain boundaries using heuristics/rules (MKDOM,
Domainer, DIVCLUS, DOMO). - Methods that rely on expert knowledge of protein
families to construct models like HMMs to
identify other members of the family (Pfam,
TigrFam, SMART). - Methods that try to infer domain boundaries by
using sequence information to predict tertiary
structure first (SnapDragon. Rigdens covariance
analysis) - Methods that use multiple alignments to predict
domain boundaries (PASS, Domination). - Others..(e.g. CSA and DGS guess based on size)
9How do you evaluate the different methods?
- No universal measures
- A variety of qualitative and quantitative
evaluation criteria, external resources and
manual analysis are used to verify domain
boundaries
10Method outline
- Source/test data SCOP
- Processed data - alignments
- Learning system
- Domain-information-content scores
- NN
- Probabilistic model
- Evaluation
- A Multi-Expert System for the Automatic
Detection of Protein Domains from Sequence
Information Niranjan Nagaragan and Golan Yona,
in the proceedings of RECOMB2003
11Overview
Intron Boundaries
DNA DATA
Seed Sequence
blast search
Sequence Participation
Multiple Alignment
Secondary Structure
Entropy
Neural Network
Correlation
Contact Profile
Physio-Chemical Properties
Final Predictions
12The source/test data set
- PDB structures with their partitions into domains
as defined in SCOP - 1ctf domain1 1-76 domain2 77-123
- Remove sequences shorter than 40 aa and almost
identical entries
13Alignments
- Search each query against a database of 1
million non-redundant sequences - Remove fragments first
- Two phase alignment procedure
- First phase blast
- Second phase multiple iteration psi-blast
- Select one representative from each group of
similar proteins - Remove proteins that are less than 90 covered
(missing information) - Number of domains ranging from 1-7
- Final set 605 multi-domain proteins and 576
single domain proteins (1/4)
14The domain-information-content of an alignment
column
- Measures that (are believed) to reflect
structural properties of proteins - A total of 20 measures
- Conservation measures
- Consistency and correlation measures
- Measures of structural flexibility
- Residue type based measures
- Predicted secondary structure information
- Intron-exon data
15Conservation measures
- Entropy some positions are more conserved than
others - Class entropy some positions have preference
towards a class of amino-acids (similar
physio-chemical properties) - Evolutionary pressure (span) sum of pairwise
similarities - Motivation consider the mutual similarity of
amino acids
16Consistency and correlation measures
- All domain appearances should maintain its
integrity - Consistency difference in sequence counts
- Asymmetric correlation consistency of individual
sequences. - Symmetric correlation reinforcement by missing
sequences - Measures are averaged over a window
17Consistency and correlation measures cont.
- Sequence termination strong but elusive
- Fragments
- Premature halt in alignment
- Loosely aligned
- Product of left and right termination scores
given c sequences that terminate at a position,
with evalues e1,e2,e3,ec
18(No Transcript)
19Measures of structural flexibility
- Indel entropy variability indicates structural
flexibility (likely to occur near domain
boundaries) - Correlated mutations indicative of contacts
Contact profiles
20Contact profile
21Residue type based measures
- hydrophobic vs. hydrophilic
- cystines and prolines
- Classes of amino acids
Predicted secondary structures
- Helices and strands are rigid
- Loops are more abundant near domain boundaries
22Intron-exon data
- Exon boundaries are expected to coincide with
domain boundaries
1
2
Protein1 Protein2 Protein3
1
2
1
3
3
2
23Score refinement and normalization
- Smoothing using a window w (optimized)
- Unification to a single scale zscore over all
positions
24Maximizing the information content of scores
- Opt for the most distinct distributions of domain
positions vs. boundary positions - Affected by the parameters (w smoothing factor)
and x (boundary window size) - Use the Jensen-Shannon divergence measure
25Examples
26- Even measures with identical distributions may be
informative in a mutli-variate model - To simplify model only the top 12 are selected
27The learning system
- A neural network is trained to model effectively
the complex decision boundary surface - Predicts correctly 94 of domain positions and
88 of the transitions in the test set - Also tried mapping from multiple positions (local
input neighborhood) to single/multiple output
28Overview
Intron Boundaries
DNA DATA
Seed Sequence
blast search
Sequence Participation
Multiple Alignment
Secondary Structure
Entropy
Neural Network
Correlation
Contact Profile
Physio-Chemical Properties
Final Predictions
29Hypothesis evaluation
- Simple model refine predictions
- Significant fraction of the positions in a window
centered at x should be predicted as transitions - Order transitions by their quality (depth of the
minima) and reject all transitions that are
within 30 residues from already predicted
transitions
30The domain generator model
- Multiple hypotheses find the best one
- Assume a model random generator that moves
repeatedly between a domain state and a linker
state and emits one domain or transition at a
time according to different source probability
distributions. - Total probability is the product
31Formally..
- S D1 D2 Dn
- We are given a sequence S (multiple alignment) of
length L and a possible partition into n domains
DD1,D2,..Dn of lengths l1,l2,..,ln (NN output) - Find the partition that will maximize the
posterior probability P(D/S) - Maximize the product of the likelihood and the
prior
32Calculating the prior P(D)
- For an arbitrary protein of length L what is the
probability to observe D - Approximate using a simplified model given the
length of the protein, the generator selects the
number of domains first and then selects the
length of one domain at a time, considering the
domains that were already generated.
33The prior probabilities
- Approximate P0(li/L) by P0(li) normalized to the
relevant range. - P0(li/L) is derived based on experimental data
34The prior probabilities (cont.)
- Calculate Prob(n/L) Prob(n,L)/P(L)
- 1
- 2
35The likelihood
- Use probabilities of observed scores considering
the two different sources - The model D partitions the sequence S into n
domains and n-1 transitions D1,T1,D2,T2,,Tn-1,Dn
that correspond to the subsequences
s1,t1,s2,t2,..,tn-1,sn - Assume domains are independent of each other
(additional test can be used)
36likelihood
- Each term P(si/Di) and P(tj/Tj) is a product over
the probabilities of the individual positions,
each one is estimated by the joint probability
distribution of the 12 features - How to estimate this probability? (independence
assumption does not hold)
37(No Transcript)
38Likelihood of individual position
- Given k random variables X1,X2,..,Xk their joint
prob. Distribution - Use first order dependencies
- For each pair, calculate the distance between the
joint prob. Distribution and the product of the
marginal distributions
39- Sort all pairs based on their dependency, and
pick the most dependent one (denoted by Y1, Y2)
and start the expansion - Select the next one based on the strongest
dependency with variables that are already in the
expansion
40- Denote by ZPILLAR(Y) the random variable that Y
is most dependent on - Of all possible dependencies involving Y3 pick
P(Y3/Z) and add it to the expansion - Proceed until you exhaust all variables
- Maximize support, minimize error
- The expansion is different for domain and
transition regions
41Finally..
- Enumerate all possible hypotheses, calculate the
posterior probability for each one, and output
the one that maximizes the prob.
42Summary of results
- Distance accuracy average distance of the
predicted transitions from their associated SCOP
transition points. - Distance sensitivity average distance of SCOP
transitions from their associated predicted
transition points. - Selectivity percentage of correct predictions
(within 10 residues from SCOP transitions) - Coverage percentage of correctly identified SCOP
transitions (within 10 residues from predicted
transitions)
43Examples
- PDB ID 2gep
- Domain Definition 8-72, 73-272, 273-352,
353-497 - Predicted Domains 1-75, 76-270, 271-352,
353-497 - PFam Definition 1-67, 273-345, 356-425
44Examples
- PDB ID 1b6s chain D
- Domain Definition 1-78, 79-276, 277-355
- Predicted Domains 1-73, 74-271, 272-355
- PFam Definition 30-167
45Examples
- PDB ID 1acc
- Domain Definition 14-735
- Predicted Domains 1-158, 159-583, 584-735
- PFam Definition 103-544
46Conclusions
- A method for predicting the domain structure of a
protein from sequence information alone - Protein/DNA data, multiple features, optimization
based on information theory principles, learning
system and final prediction using the
domain-generator model (with confidence values). - Exhaustive hypothesis evaluation
- Fully automatic and fast
- Perform very well even compared to the best
manual and semi-manual methods out there (also on
CATH data) - Dare to say can be used to verify domain
assignments based on structural data - Improvements other learning systems, more
features
47Acknowledgments
- Niranjan Nagarajan
- SCOP
- CATH
- PSI-BLAST
- Pfam
- InterPro
- NSF