Download Lecture 4 - Multiple Sequence Alignment | CMSC 828T and more Study notes Computer Science in PDF only on Docsity!
CMSC 838T โ Lecture 4
CMSC 838T โ Lecture 4
X Multiple sequence alignment (MSA)
0 Alignment containing multiple DNA / protein sequences 0 Look for conserved regions โ similar function
Multiple Sequence Alignments - Motivation
X Identify highly conserved residues
0 Likely to be essential sites for structure / function 0 More precision from multiple sequences 0 Better structure / function prediction, pairwise alignments
X Building gene / protein families
0 Use conserved regions to guide search
X Basis for phylogenetic analysis
0 Infer evolutionary relationships between genes
X Develop primers & probes
0 Use conserved region to develop O Primers for PCR O Probes for DNA microarrays
CMSC 838T โ Lecture 4
Multiple Sequence Alignment (MSA)
X Outline
0 Basic concepts & terms 0 Global alignment O Optimal โ dynamic programming (MSA) O Progressive โ pairwise (PILEUP, CLUSTALW) O Iterative progressive (MULTALIN) O Block-based (DIALIGN) 0 Local alignment (motif finding) O Patterns (MOTIF, PROTOMAT) O Statistical profiles (HMMER2, PSI-BLAST) 0 Viewing & editing multiple sequence alignments
Terminology (for Proteins)
X Family
0 Group of proteins of similar biochemical function with (roughly) > 50% sequence identity when aligned 0 Family is transitive, even if sequence identity < 50% O A โ B and B โ C implies A โ C 0 1940 protein families in Protein DataBank (v1.61, Nov 2002)
X Superfamily
0 Group of protein families related by distant yet detectable sequence similarity 0 1100 protein superfamilies in Protein DataBank (v1.61)
CMSC 838T โ Lecture 4
Multiple Sequence Alignment - Block
X Ungapped conserved sequence pattern
X Types of blocks
0 Exact โ composed of identical segments 0 Uniform โ found in every sequence 0 Consistent โ can be part of global MSA
A A T G T G T G A G A C T C
C T T A G T G T A C C A C G
A A C A T G T A T A C G T A C G
A C G T G C C T A C T A
non-consistent
Diagonal
MSA - Global vs. Local Alignment
X Protein structure
0 Derived from a limited number of building blocks (domains) that have been mixed and shuffled through evolution 0 Proteins can thus share a global or local relationship
X Global sequence alignment
0 Alignment over entire sequence (near same length)
X Local sequence alignment
0 Alignment over parts(s) of sequence
Global Alignment Local Alignment
CMSC 838T โ Lecture 4
Multiple Sequence Alignment - Approaches
X Progressive global
alignment
0 If sequences related over entire length
X Block-based global
alignment
0 If related by large consistent blocks
X Local alignment
0 If related by small non-consistent blocks
Multiple Sequence Alignments - Issues
X No single โcorrectโ answer
0 Ideally, find single evolutionary correct alignment 0 In practice, evolutionary history must be inferred 0 Try find sequence alignment โ good structure alignment
X Typical alignment target
0 Protein family with ~30% identity
X Computationally expensive
0 โOptimalโ solution exponential in number of sequences 0 Greater reliance on (greedy) heuristics
X Benefits from user interaction
0 Select which sequences to include in alignment 0 Select which regions to align 0 Edit resulting alignments
CMSC 838T โ Lecture 4
Global MSA - Approach
X General approach to global MSA
- Find sequences to align (e.g., result of pairwise search)
- Locate region(s) of similar length to include in alignment
- Apply global alignment algorithm
- Refine alignment (repeat as needed)
- Inspect resulting alignment 0 Identify conserved physical / chemical properties
- Remove seriously misaligned sequences
- Reapply algorithm
- Add back remaining sequences 0 While preserving key features of alignment
- If looking for local alignment, reduce sequence length to highly conserved regions, align to conserved region
Global MSA - Dynamic Programming (DP)
X โOptimalโ dynamic programming [Sankoff+ 1983]
0 Assume k sequences of length n 0 Attempt to maximize sum-of-pairs (SP) score 0 Build F, a k-dimensional table of length n+1 (n k^ elements) 0 Recursive formula โ F( i ) = max( F(i-1) + SP(columni ) )
X Complexity
0 O(nk^ ) entries to fill 0 Each entry combines O(2k^ ) other entries 0 Total cost = O(2k^ nk)
X Bounded search (MSA) [Carillo & Lipman 1988]
0 Apply heuristic aligment, use resulting SP to bound search 0 Significant speed improvement, still limited to small values of k
K=
CMSC 838T โ Lecture 4
Global MSA โ Progressive Global Alignment
X Motivation
0 Reduce cost by building global alignment incrementally
X Approach
- Compute distance between all pairs of sequences
- Build simple guide tree reflecting distance between sequences 0 Use UPGMA (PILEUP) OR neighbor-joining (CLUSTALW)
- Align sequences following guide tree, starting at leaves 0 Align consensus sequences OR profiles 0 Use optimal or heuristic pairwise algorithms 0 Attempt to place gaps between conserved regions
X Problems
0 Greedy approach dependent on initial pairwise alignments 0 Cannot fix early mistakes (gaps cannot be removed)
Global MSA โ CLUSTALW
X Algorithm [Thompson+ 1994]
0 Calculate evolutionary distances from alignment scores 0 Performs pairwise alignment of profiles (probabilities of residues at each position) using dynamic programming 0 Later calculates consensus sequence from profile
X Heuristics for improving multiple alignments
0 Weight sequences to compensate for biased representation 0 Scoring matrix chosen based on expected similarity from tree O E.g., nearby โ BLOSUM 80, distant โ BLOSUM 50 0 Gap penalty modified by residue (function) at position O E.g., Higher gap penalty for hydrophobic residues 0 Gap penalty higher if first gap in column & nearby gaps 0 Dynamically adjust guide tree to defer poor alignments
CMSC 838T โ Lecture 4
Multiple Sequence Alignment (MSA)
X Outline
0 Basic concepts & terms 0 Global alignment O Optimal โ dynamic programming (MSA) O Progressive โ pairwise (PILEUP, CLUSTALW) O Iterative progressive (MULTALIN) O Block-based (DIALIGN) 0 Local alignment (motif finding) O Patterns (MOTIF, PROTOMAT) O Statistical profiles (HMMER2, PSI-BLAST) 0 Viewing & editing multiple sequence alignments
Local MSA (Motif-finding) - Approach
X Motivation
0 Find local regions of high similarity (motifs) 0 Align based on motifs
X Approach
0 Find motifs O Patterns O Blocks O Statistical profiles X Position-specific scoring matrix (PSSM) X Hidden Markov model (HMM) 0 Align sequences O Preserve motifs as much as possible
CMSC 838T โ Lecture 4
Terminology (for Protein Sequences)
X Pattern
0 Deterministic syntax describing well-conserved region
X Profile
0 Probabilistic syntax describing well-conserved region 0 Score-based representations O Position-specific scoring matrix (PSSM) O Hidden Markov model (HMM)
X Pattern & profile
0 Can be used to search for motifs / domains of biological significance that characterize protein family
Significance of Patterns / Motifs
X DNA
0 Recognition sites of restriction endonucleases 0 Codons specifying the amino acid sequence of a protein 0 Intron splice sites 0 Promoter 0 Binding sites for regulatory proteins which activate or repress transcription
X Proteins
0 Presence of active sites 0 Prediction of protein secondary structure 0 Presence of signals used to localize the protein in the cell
CMSC 838T โ Lecture 4
Local MSA - Statistical Profile
X Position-specific scoring matrix (PSSM)
0 Summary representation for (aligned) conserved region 0 Stores probability of element at each position in sequence 0 Entries usually stored in log-odds form 0 Weight entries by 1) average proportion, 2) evolutionary dist. 0 Consensus โ most likely base / residue at each position
A A C
A C G
A - T
A A T
3) T 0.00 0.25 0.25 0.50 -
2) A 0.75 0.25 0.00 0.00 -
1) A 1.00 0.00 0.00 0.00 -
Cons A C G T Gap
Local MSA - Statistical Profile
X Hidden Markov model (HMM)
0 Statistical summary representation for conserved region 0 Model stores probability of match, mismatch, insertions, and deletions at each position in sequence 0 Alignment of conserved region not necessary, but helpful
A A C
A C G
A - T
A A T
match states
insert states
delete states
b e g i n
e n d
A C G T
A C G T
A C G T
CMSC 838T โ Lecture 4
Local MSA - HMMs
X HMM construction (HMMER2, PSI-BLAST)
- Initialize model with estimated amino acid transition probabilities, using PAM / BLOSUM / Dirichlet mixtures
- For each โtrainingโ sequence containing conserved region O Find all possible paths for sequence through model Forward-backward algorithm = O(size ร sequences) O Increase weight (probability) of each path taken
X HMM properties
0 Ideally use 20-100 training sequences to build model O With better initialization, smaller โtraining setโ sufficient 0 Well grounded in probability theory (statistical significance) 0 Explicit gap penalties not needed (automatically trained) 0 Can extract consensus sequence (w/ dynamic programming)
Calibrating Profiles for PSSM & HMM
X Profiling methods (PSSM, HMM)
0 Training set used to build profile may be biased / skewed O Over-represented sequences (common motifs) O Under-represented sequences (rare residues) 0 Resulting profile matches training set, not desired motif
X Weighting / calibration
0 Differentially weight sequences to compensate for non- representative sampling in training set 0 Similar sequences โ lower weights 0 Rare sequences โ higher weights 0 Maximum discrimination โ set of weights that best differentiate between real matches and background noise
X Simulated annealing to avoid local maxima
CMSC 838T โ Lecture 4
Searching Based on MSA Profile
X Advantages
0 Searches based on domain, not sequences 0 Greatly improved sensitivity in practice
X Dependent on user selection / deletion of sequences
0 Once included in profile, sequence will score well 0 Including false positives (mismatches) reduces accuracy O Can use pairwise alignment to compare sequences O Demonstrate sequence can mutate into other sequences
Multiple Sequence Alignment (MSA)
X Outline
0 Basic concepts & terms 0 Global alignment O Optimal โ dynamic programming (MSA) O Progressive โ pairwise (PILEUP, CLUSTALW) O Iterative progressive (MULTALIN) O Block-based (DIALIGN) 0 Local alignment (motif finding) O Patterns (MOTIF, PROTOMAT) O Statistical profiles (HMMER2, PSI-BLAST) 0 Viewing & editing multiple sequence alignments
CMSC 838T โ Lecture 4
Viewing & Editing Multiple Alignments
X Motivation
0 Use multiple sequence alignment as starting point 0 Improve usability / readability O Format alignments O Add annotations 0 Improve alignments manually with expert knowledge O Find biologically significant regions
X Multiple sequence alignment tools
0 Viewers O ClustalX, Jalview, Cinema, Sequence logos 0 Editors / annotation O SeqVu, MACAW
Viewing Multiple Sequence Alignments
X Coloring scheme
0 Helps better visualize conserved regions
X Example color code
0 AVFPMILW: RED, Small O (small + hydrophobic (including aromatic -Y)) 0 DE: BLUE, Acidic 0 RHK: MAGENTA, Basic 0 STYHCNGQ: GREEN, Hydroxyl + Amine + Basic โ Q 0 Others: Grey
Multiple Sequence Alignment (MSA)
X Summary
0 Many multiple sequence alignment algorithms 0 Most global alignment algorithms too expensive O Exception - progressive pairwise alignment (heuristic) 0 Local alignment algs. try to find essential conserved regions O Can be very simple (matching motifs) O Or use heavy-duty statistical analysis models 0 Searches using MSA more sensitive than pairwise alignments 0 When using MSA to search / edit motifs O Knowledge of biochemistry provides major advantage
