The Immune Epitope Database Analysis Resource: MHC class I peptide binding predictions. Edita Karosiene, Ph.D.

Transcription

1 The Immune Epitope Database Analysis Resource: MHC class I peptide binding predictions Edita Karosiene, Ph.D. edita@liai.org IEDB Workshop October 29, 2015

2 Outline Introduction MHC-I peptide binding prediction tool Web user interface Guidelines on choosing Prediction methods Prediction thresholds HLA alleles Exercises 1

3 Introduction 2

4 MHC class I molecules Expressed by almost all nucleated cells and present antigen to CD8 + T cells The binding groove is closed at both ends, can accommodate peptides of 8-15 AA Only α chain impacts binding. α chain β2-microglobulin HLA-A*02:01, PDB entry 1DUZ 3

5 MHC binding predictions MHC molecules are highly polymorphic thousands of different variants exist Experimental characterisation of peptide MHC interactions is highly costintensive Prediction methods facilitate selection of potential epitopes from a pool of peptides Peptide binding data Machine learning algorithms PEPTIDE IC50 (nm) AAAAAGTTV AAAAAVAAE 211 AAAAAVAAE 5 AAAAPYAGW AAACAGTTV 6 AAADAGTTV 6 AAADLFRGT 8791 AAADLFRGV 1773 AAADLFRGW Artificial Neural Networks (ANN) 4

6 Binding data for MHC class I 126 MHC molecules: 83 human: HLA-A, HLA-B and HLA-C 18 macaque: Mamu-A, Mamu-B 8 chimpanzee: Patr-A, Patr-B 6 mouse: H-2 Db, Dd, Kb, Kd, Kk, Ld 7 cattle: BoLA-HD6 3 pig: SLA1, SLA2, SLA3, SLA6 1 rat: RT1A Predictors were re-trained in the middle of year

7 MHC-I peptide binding prediction tool Web user interface 6

8 MHC-I binding predictions on IEDB home page 7

9 MHC-I binding predictions 8

10 Instructions 9

11 Examples 10

12 References 11

13 Download Peptide binding data 12

14 Contact 13

15 Submission Submit sequence: Space separated sequences One continuous sequence FASTA format 14

16 Submission Choose file with sequence(s) 15

17 Submission Choose sequence format "auto detect format": FASTA if ">" is found Continuous sequence otherwise 16

18 Submission Choose prediction method IEDB Recommended: Consensus if predictions available for a chosen MHC molecule NetMHCpan, otherwise Consensus combines ANN, SMM, CombLib 17

19 Submission Choose species Specify MHC allele(s) and peptide length Format: HLA-A*02:01,9 HLA-B*15:01,9 HLA-A*02:06,10 18

20 Submission Choose output sorting method 19

21 Submission Optional filtering of results 20

22 Submission Choose output format Submit 21

23 Prediction results A percentile rank score for a peptide is a percentage of randomly sampled peptides scoring better than the peptide. 22

24 Prediction results Individual scores for different methods 23

25 Prediction results SMM method was chosen 24

26 How to choose prediction method 25

27 Groups of prediction methods Allele-specific methods Training data: peptide sequences binding affinities to one specific MHC molecule Predictions: molecules from the training set. Stabilized Matrix Method (SMM) SMM PMBEC CombLib NetMHC/ANN Consensus Pan-specific methods Training data: peptide sequences MHC sequences binding affinities Predictions: any known MHC molecule. NetMHCpan PickPocket NetMHCcons 26

28 Consensus Motivation: combined predictions from different predictors may result into higher predictive performance than any of the individual predictors Consensus Combines NetMHC, SMM and CombLib Requires that all methods give predictions on the same scale percentile ranks are used PMID: CombLib = 0.2 SMM = 0.4 ANN = 0.7 MEDIAN Consensus =

29 Recommendations IEDB Recommended Addresses the problem of choosing from many available prediction methods Uses the best method for a given MHC molecule Consensus if predictions for NetMHC, SMM or CombLib are available NetMHCpan, otherwise Benchmark of the methods: Consensus > NetMHC > SMM > NetMHCpan > CombLib Recommendation may change with the new benchmark studies 28

30 How to choose prediction threshold 29

31 Two main choices for binders 1. Binding affinity (IC50) threshold of 500 nm identifies peptide binders recognized by T cells IC50 < 50 nm high affinity IC50 < 500 nm intermediate affinity IC50 < 5000 nm low affinity Sette et al. 1994, J. Immunology (PMID: ) Ensures that all peptides have reasonable affinity (at lest intermediate affinity). 1. Pick top 1% of peptides for each (MHC, length) combination to cover most of immune responses Moutaftsi et al. 2006, Nat. Biotech (PMID: ) Kotturi et al. 2007, J. Virology (PMID: ) Ensures equal number of peptides per allele. 2. Percentile rank score threshold of

32 Different peptide-binding repertoires The size of the peptide repertoire binding at a given affinity varies between alleles. HLA-A*02:01 HLA-A*01:01 Allele-specific affinity cutoffs Paul et al. 2013, J. Immunol. (PMID: ) 31

33 Allele-specific thresholds 32

34 Recommendations Both approaches (affinity and ranking) are reasonable, and have been applied in numerous studies Cut-offs can be combined (peptides in top 1% and IC50 < 500 nm) Current studies suggest that allele specific thresholds can be derived. 33

35 How to choose representative alleles 34

36 MHC polymorphism Known HLA protein molecules: HLA-A 2,041 HLA-B 2,668 HLA-C 1,677 Total: 6386 We have peptide binding data for 80 => 1.3 % 35

37 How to deal with MHC polymorphism Frequency of alleles is changing across different populations worldwide MHC alleles can be classified into relatively few supertypes based on their peptide binding similarities Sette and Sidney, 1998, Curr. Opin. Immunol. (PMID: ) Sette and Sidney, 1999, Immunogenetics (PMID: ) Goulder & Watkins, 2008, Nature Reviews Immunology 36

38 Recommendations Reference set of 27 alleles has been assembled covers > 97% of population HLA-A Frequency HLA-B Frequency A*01: B*07: A*02: B*08: A*02: B*15: A*02: B*35: A*03: B*40: A*11: B*44: A*23: B*44: A*24: B*51: A*26: B*53: A*30: B*57: A*30:02 5 B*58: A*31: A*32: A*33: A*68: A*68:

39 Recommendations 38

40 Exercise 1 finding Malaria minimal epitope 39

41 Exercise 1 finding Malaria minimal epitope Background Apical membrane antigen-1 (AMA1) is a protein expressed in the membrane of Plasmodium falciparum in sporozoite, liver and blood stages. AMA1 has been tested in clinical trials to give both CD4 + and CD8 + responses and is considered as a good candidate for a multi-antigen malaria vaccine. Methods Five volunteers were immunized with a vaccine containing full length of AMA1. A peptide pool of 15-mers overlapping by 11 amino acids in the AMA1 sequence was constructed. ELISpot responses of the peptides from the peptide pool were tested among the volunteers. HLA typing was done for each volunteer. Problem statement Use IEDB prediction tools to determine the best epitope within a 15-mer: LLSAFEFTYMINFGR. Volunteer s HLA set: HLA-A*02:01, HLA-A*26:01, HLA-B*18:01, HLA-B*44:02. Reference Sedegah M. et al. 2010, Malaria Journal (PMID: ) 40

42 Exercise 2 choosing epitopes for a global Dengue virus vaccine 41

43 Exercise 2 choosing epitopes for a global Dengue virus vaccine Imagine that you want to design a vaccine for Dengue virus that would work across the world. Use Dengue virus 1 isolate from Sri Lanka as a reference sequence (GenBank: JN ). You would have to use HLA allele reference set (27 alleles) to make predictions. However, due to the limited time choose only 1 allele from the reference list: HLA-A*01:01. Answer the questions: 1) How many peptides would you choose if you follow top 1% rule? 2) How many peptides would you choose if you only want to have peptides with a percentile rank score < 1.0? 3) How many of the top 10 peptides are known as Dengue epitopes in the IEDB data base? 42

44 Thank you for your attention! Questions?