Protein families, domains and motifs for functional prediction

1. Protein families, domains and motifsfor functional prediction June 27, 2019

2. Outline • Usefulness of protein domain analysis • Types of protein domain databases • SMART, HMMER and Interpro protein domain database • Uniprot protein annotation • Predicting post-translational modifications

3. Protein families • Groups of homologous sequences (within and across species) that share similar functions and domains • Examples: • Carbonic anhydrases (14 in humans) • Chitin synthases (8 in C. neoformans) • Ser/Thr kinases

4. Protein domains • Conserved part of protein sequence that can evolve, function and exist independent of the rest of the protein chain • Often independently stable and folded • Can recombine or evolve from gene duplications into proteins with different combinations of domains

5. Protein motifs • Short linear peptide sequences that serve a specific function for the protein, but will not be stable or fold independent of the rest of chain • Protein-protein interaction, ligand interactions, cleavage sites, targeting • Examples: • 14-3-3: Interaction with kinases • KELCH: ubiquitin targeting • SUMO: site recognized for modification by SUMO • Often found within intrinsically disordered regions

6. Predicting function for unknown proteins • Do they belong (by sequence homology) to a protein family? • Do they contain known protein domains? • Do they have motifs that suggest a specific function?

7. Limitations of annotation • Even in a model organism with large amount of resources, most genes are still annotated by similarity • Often, the name given is based on the BEST match to a particular domain or known protein • But…

8. Limitations of BLAST • Likelihood of finding a homolog to a sequence: • >80% bacteria • >70% yeast • ~60% animal • Rest are truly novel sequences • ~900/6500 proteins in yeast without a known function • NAME: Similar to yeast protein YAL7400 not very informative

9. Limitations of similarity • Proteins with more than one domain cause problems. • Numerous matches to one domain can mask matches to other domains • Increased size of protein databases • Number related sequences rises and less related sequence hits may be lost • Low-complexity regions can mask domain matches

10. Proteins are modular • Individual domains can and often do fold independently of other domains within the same protein • Domains can function as an independent unit (or truncation experiments would never work) • Thus identity of ALL protein domains within a sequence can provide further clues about their function

11. Proteins can have >1 domain The name: protein kinase receptor UFO doesn’t necessarily tell you that this protein also contains IgG and fibronectin domains or that it has a transmembrane domain

12. Domains are not always functional • If a critical residue is missing in an active site, it’s not likely to be functional • A similarity score won’t pick that up

13. Protein signature databases • Identify domains or classify proteins into families to allow inference of function • Approaches include: • regular expressions and profiles • position-specific scoring matrix-based fingerprints • automated sequence clustering • Hidden Markov Models (HMMs)

14. PROSITE • Regular expression patterns describing functional motifs M-x-G-x(3)-[IV]2-x(2)-{FWY} • Enzyme catalytic sites • Prosthetic group attachment sites • Ligand or metal binding sites • Either matches or not • Some families/domains defined by co-occurrence

15. Citrate synthase G-[FYAV]-[GA]-H-x-[IV]-x(1,2)-[RKTQ]-x(2)-[DV]-[PS]-R

16. Profile-HMMs • Models generated from alignments of many homologues then counting frequency of occurrence for each amino acid in each column of the alignment (profile). • Profile-HMMs used to create probabilities of occurrence against background evolutionary model that accounts for possible substitutions. • Provides convenient and powerful way of identifying homology between sequences. • Find domains in sequences that would never be found by BLAST alone

17. HMM domain databases • PFAM • Classify novel sequences into protein domain profiles • Most comprehensive; >16,000 protein families (v31) • Now working with Uniprot • SMART • Signaling, extracellular and chromatin proteins • Identification of catalytic site conservation for enzymes • TIGRFAMs • Families of proteins from prokaryotes • PANTHER • Classification based on function using literature evidence

18. PFAM • Manually curated profiles • a statistical measure of the likelihood that an alignment occurred by chance alone • Does not indicate functionality

19. PFAM Summary

20. PFAM Domain Organization

21. PFAM “parts” definitions 1) Domain: A collection of related sequence regions that form a distinct structural unit. 2) Family: A collection of related sequence regions that may contain one or more domains, but where there is insufficient evidence to support subdivision. 3) Repeat: A short unit which is unstable in isolation but forms a stable structure when multiple copies are present. 4) Motif: A short unit that carries a distinct role, for example in metal binding. 5) Coiled-coil: Regions of a protein that form alpha-helices that align against each other to form a distinctive structure called a coiled-coil. 6) Disordered: Regions on proteins that are inherently disordered but have sequence conservation.

22. SMART database • SMART: Simple Modular Architecture Research Tool • Focus on signaling, extracellular and chromatin-associated proteins • Curated models for >1300 domains • Use? • I have several kinase domains in my protein list and want to know which ones are functional. • What other signaling proteins are in my list? • What other domains are found in signaling proteins?

23. SMART: Search interface Uniprot or Ensemble Protein Accession number Add other searches

24. SMART Output

25. HMMER • Fast, sensitive protein homology searches using HMMs • Results include taxonomic distribution of matches • PFAM domains • Transmembrane, coiled-coil, signal peptide and intrinsically disordered regions • Typically faster* than BLAST or InterProScan * PHMMER search speeds vary depending on usage

26. ADCYC3 PHMMER search

27. InterPro Scan • Combines search methods from several protein databases • Uses tools provided by member databases • Uses threshold scores for profiles & motifs • InterPro convenient means of deriving a consensus among signature methods • InterPro records integrated with Uniprot. • Slow to return search results • Can bookmark it and come back to it later

28. ADCYC3 Interpro match Different Domain databases: PF -> PFAM PS -> PROSITE PR -> PRINTS SM -> SMART Group different domains into related signature

29. Adenylyl cyclase class-3/4guanylyl cyclase

30. Function from sequence • Membrane bound or secreted? • GPI anchored? • Cellular localization? • Post-translational modification sites?

31. CBS prediction services • Protein sorting • SignalP, TargetP, others • Post-translational modification • Acetylation, phosphorylation, glycosylation • Immunological features • Epitopes, MHC allele binding, ect • Protein function & structure • Transmembrane domains, co-evolving positions

32. Transmembrane domain prediction

33. Phosphorylation prediction

34. O-glycosylation

35. EMBOSS Open source software for molecular biology • Predict antigenic sites • Useful if want to design a peptide antibody • Look for specific motifs, even degenerate • Known phosphorylation motifs • Find motifs in multiple sequences with one submission • Get stats on proteins/nucleic acid sequences • Sequence manipulation of all kinds

36. Today in lab • Tutorials on protein information sites • Create two sublists with DAVID • Obtain protein accession numbers for the cluster using Uniprot • Submit to SMART database to characterize/analyze the domains for signaling proteins • Pick 2 proteins to do additional predictions