Protein Sequence Analysis. BME 110: CompBio Tools Todd Lowe April 19, 2007 (Slide Presentation: Carol Rohl)

Transcription

1 Protein Sequence Analysis BME 110: CompBio Tools Todd Lowe April 19, 2007 (Slide Presentation: Carol Rohl)

2 Linear Sequence Analysis What can you learn from a (single) protein sequence? Calculate it s physical properties Molecular weight (MW), isoelectric point (pi), amino acid content, hydropathy (hydrophilic v. hydrophobic regions) Does not take into account post-translational modifications of protein, so are usually not 100% accurate Identify sequence motifs and families Signal sequences, transmembrane domains, coiled-coils, posttranslational modification sites, secondary structure (nonhomologous) Domains, functional motifs (homologous)

3 3-D Structure Analysis Visualization Domain structure, global fold, active sites, point mutations, SNPs, splice sites Evaluate structure quality Calculate physical properties Surface areas, distances, side-chain conformations, contact maps Structural alignment (ie similarity to other structures) Prediction Physical properties: binding affinity, pka s stability, specificity 3D structure (homology modeling, fold recognition, de novo) Advanced: protein design, docking of two proteins, active site modeling

4 Sequence Databases SwissProt (ExPASy) Highly curated, updated less frequently TrEMBL (ExPASy) Translated nucleotide sequences Automatic translation, fast but less info UniProt (EBI) Unified Protein Resource Combines SwissProt, TrEMBL, PIR sequences

5 Sequence Analysis Sites For protein sequences and tools to analyze them, the two major centers are: ExPASy : Expert Protein Analysis System Many tools: Databases: SwissProt, TrEMBL NCBI : Entrez Protein and Domains PIR: Protein Information Resource (folded into UniProt consortium; no longer major resource site)

6 More Sequence Databases Non-redundant NR (NCBI), UniRef (PIR/EBI) Reference RefSeq (NCBI) re-annotated by NCBI Domains/Families Pfam protein families (Sanger Center + 4 mirror sites) SMART Simple Modular Architecture Research Tool CDD Conserved protein Domain Database (NCBI), combines Pfam, SMART, and COGs databases InterPro (based on UniProt, at EMBL-EBI) Many others

7 Structure Databases Experimental: PDB: Protein Data Bank Families: SCOP, CATH, Dali database, Homstrad Models/Predictions ModBase SwissModel NOTE: All these databases are described in January Database issue of Nucleic Acids Research (plus other kinds of databases). Also, links to them

8 Protein Sequence Analysis Tools ExPASy Proteomics Tools Calculate physical properties Predict sequence motifs what ExPASy calls Topology : localization, TM domains Signal sequences, postranslational modifications Search pattern and profile collections PredictProtein and Meta-PP A meta-server providing access to many servers with one submission form

9 Secondary Structure Prediction Three good methods: Psipred Sam-T02/T04/T06 PhD (PredictProtein) Compare a couple methods Use the three-state predictions

10 SEQUENCE <--> STRUCTURE <--> FUNCTION Evolutionary selection operates on function Structure is more closely linked to function than is sequence, so structure tends to be more conserved than sequence. Need to search farther in sequence space to find proteins with related structures and functions.

11 Detecting Remote Similarities Remote similarities can more easily be detected by comparing protein sequences DNA sequences change faster than protein sequences (wobble position, redundant codons) 4 letter DNA code vs. 20 letter amino acid code means that matches by chance are more likely in DNA; The protein code has more information in it!

12 Detecting Homology NEAR Evolutionary Distance FAR DNA Sequence Protein Sequence SIMILARITY Protein Structure BLASTn BLASTp PSIBLAST Fold Recognition METHODS

13 Similar Sequences Share Similar Structures Compare all pairs of proteins in the same family (pairs for which homology is very probable) Homologs do not necessarily share much sequence similarity. Proteins with >30% sequence identity almost always share the same fold More structurally similar All other immunoglobulins Sauder et al., (2000) Proteins 40:6

14 BLASTP Heuristics Most sequences will be unrelated Related sequences are likely to have short stretches of identities Use identical (or closely related) short words as seeds for local alignment Like BLASTn with shorter words Will return global alignment if proteins similar over whole length

15 BLASTp Scoring Matrices less divergent BLOSUM 80 BLOSUM62 BLOSUM45 PAM1 PAM120 PAM250 more divergent BLOcks amino acid SUbstitution Matrices Calculated directly from substitution frequencies in local, ungapped alignments of biochemically related sequences Number indicates the highest sequence similarity between sequences used. Percent Accepted Mutations Derived from global alignments of closely related sequences (85% identity) using an evolutionary model to extrapolate to lower identities Number indicates evolutionary distance If in doubt, use BLOSUM. More suited to searching databases using local alignment. No assumed model of evolutionary divergence.

16 Other BLASTp Parameters Gap penalties The harder it is to open/extend a gap, the fewer will be made. If you re looking for close sequences, gap penalties should be higher. Databases NR (non-redundant, translated gene sequences) SwissProt PDB Phylogenetically specific (i.e. Archaea only)

17 PSIBLAST Position-Specific Iterated BLAST Creates a scoring matrix specialized for your sequence Allows more distantly related sequences to be identified Steps 1. Use BLASTp and identify related sequences (E-value threshold) 2. Create a profile from related sequences 3. Search for related sequences using this profile 4. Repeat

18 BLASTing The Protein Universe BLOSUM45 BLOSUM80

19 Evolution And The Protein Universe

20 PSIBLASTing The Protein Universe Iteration 1 Iteration 2

21 Sequence Profiles Align all sequences and count how often each amino acid occurs at every position. Combine with prior information about substitution frequencies using pseudocounts from BLOSUM62 Convert to log odds score to give a Position-Specific Scoring Matrix (PSSM)

22 A Sample PSSM A R N D C Q E G H I L K M F P S T W Y V 1 M K W V W A L L L L A A W A A A S G T W Y A From Pevsner, Bioinformatics and Functional Genomics (ISBN ) John Wiley & Sons, Inc.

23 PSSM Corruption False positives can occur in a PSIBLAST search of the PSSM becomes corrupted. How do PSSMs become corrupted? One sequence that is not homologous to the query gets included in the alignment used to make the PSSM. The PSSM now looks a bit like this spurious sequence and will match well to other similar spurious sequences. These additional spurious sequences that are detected are included in the new alignment, amplifying the corrupting signal. Once a bad sequence is included in the PSSM, the search veers of course, and cannot be corrected.

24 Preventing PSSM Corruption Apply filtering of biased composition regions. (Low complexity filter) Use better methods to estimate the E-value (compositionbased statistics) Increase threshold for judging two sequences to be similar: adjust E value from (default) to a lower value such as E = Manually inspect the output from each iteration and remove suspicious hits.

25 PHI-BLAST: Pattern-Hit-Initiated BLAST Combines matching of regular expressions with local alignments surrounding the match. What other proteins contain a particular sequence pattern and are similar in the vicinity of this pattern? May filter out cases where pattern matches randomly and doesn t indicate homology Pattern matching uses ScanProsite syntax Sequence similarity search is like PSIBLAST

26 Syntax Rules for Patterns [ ] any one of the listed characters allowed { } any character except the listed ones allowed x(n) n positions in which any residue is allowed x(n,m) n to m positions in which any residue is allowed (n,m) Examples: GXW[YF][EA][IVLM] Matches: GTWFEL GKWYAI Does not match: GGWYFEI GWYEI E[LIV]X(0,3)PP[STG] Matches: ELPPS ELPPPSTG EVIPPG Does not match: ELIVPPPPG

27 Gene Discovery with BLAST Start with the sequence of a known protein tblastn Search a DNA database (e.g. HTGS, dbest, or genomic sequence from a specific organism) inspect Search your DNA or protein against a protein database (nr) to confirm you have identified a novel gene blastx or blastp nr Find matches [1] to DNA encoding known proteins [2] to DNA encoding related (novel!) proteins [3] to false positives