*Watch the video at the end of the presentation Seminar led by Dr. Xavier de la Cruz, ICREA Research Professor. Head of the Translational Bioinformatics in Neuroscience group of VHIR, at VHIR (22nd November 2012). Content: The need to identify the pathological character of mutations may arise in different contexts in biomedical research. However, the methods available to address this problem essentially depend on the number of cases under analysis. When we work with only a few mutations we can use an artisan-like approach, where all information available on protein sequence, structure and function is manually retrieved and studied. However, when we need to characterize many variants, as can be the case in exome projects, faster methods are required to assess their pathogenicity. In my talk I will illustrate the principles underlying these two approaches with examples from the study of Fabry disease mutations, resulting from our collaborative work at the VHIR.

1. Identification of pathological mutations
from the single-gene case to exome
projects: lessons from the Fabry disease
Xavier de la Cruz

2. Identificationofpathological…
 Interpretationcontextofmutation data
 Identifyingpathologicalmutations:
◦ Presenttools
◦ Problems?
 A VHIR-basedtoolformutationscoring
◦ Development
◦ Performance
 Futuredirections
◦ Implementing a
standardizedmutationreport

3. From base pairs to bedside
(Green & Guyer, Nature, 2011)
Understanding Understanding Improving
genome disease biology healthcare
structure Understanding Advancing effectiveness
genome biology medical science
1990-2003
Genome
Project
2004-2010
2011-2020
Beyond 2020

5. So, is there a problem?
2017
$517000
$285000
INTERPRETATION COST
<$100

6. From base pairs to …
Sample
Exome sequencing
Variant identification
and
quality control
INTERPRETATION

7. The interpretation problem
 “…enormous amounts of raw data, but
still very little understanding of what it
means”
 Exome sequencing context:
◦ Identify disease causative variants
◦ Prioritize of variants
◦ Speed: can we do this for 100‟s-1000‟s
variants in “reasonable” time?
◦ Reliability: can we provide good error models
for counseling/diagnosis/prognosis?

8. Exome-ready mutation annotation
tools
 PolyPhen (Adzhubei et al., 2010), SIFT
(Kumar et al., 2009):
◦ Mutation mining
 pathological : databases + literature + private
datasets
 neutral: experimental sets (LacI, lysozyme,
etc), evolutionary model, databases (dbSNP)
◦ Building model: machine learning

9. PERFORMANCE

10. CONDEL
González-Pérez & López-Bigas, 2011
ROC area:
•CONDEL: 0.849
•CAROL: 0.852
CAROL
Lopes et al., 2012

11. Limitsofpresentannotationtools
 Consensustools:
◦ Understandingof molecular
damageisharder
◦ Theydependontheexistenceofprimarytoold
s (PolyPhen, SIFT)
 Primary (PolyPhen, SIFT):
◦ Average overmanymutationsand genes

12. Type III Hereditary Hemochromatosis – TFR2
• TFR2, a dimeric type II transmembrane membrane protein expressed mostly in
the liver and CD71+ early erythroids.
• At least 50 families and 69 patients have been described with mutations in
TFR2 gene.

14. Fabry disease
 Systemic disorder characterized by:
progressive renal failure,
cardiovascular or cerebrovascular
disease, etc.
 Caused by mutations in lysosomal
enzyme -galactosidase A

15. CYS52

16. PRESENT PREDICTORS
GENE 1 PATHOL. MUT. 1
GENE 2 PATHOL. MUT. 2
MUTATION
GENE 3 PATHOL. MUT. 3 PREDICTOR
……… ………

17. GENE-SPECIFIC
PREDICTORS
GENE 1 PATHOL. MUT. 1 MUTATION PREDICTOR 1
GENE 2 PATHOL. MUT. 2 MUTATION PREDICTOR 2
GENE 3 PATHOL. MUT. 3 MUTATION PREDICTOR 3
……… ……… MUTATION PREDICTOR …

18. Improving mutation annotation tools
 Train in single genes (Ferrer-Costa et
al., 2004): increase 5%-10%
successrate

19. METHOD
S

20. MUTATION
property 1 property N
property 2 property i property j
PATHOLOGICAL / NEUTRAL

21. DATAMINE pathological mutations
CHARACTERIZE PROTEIN DAMAGE
BUILD COMPUTATIONAL MODEL
•Experimental
Application:
study
• score
•Counseling
• prioritize
•Etc

22. Datamine Pathological
Mutations
 General databases: UniProt, OMIM
 Specificdatabases:
◦ Fabrydatabase(http://fabry-database.org/)
◦ p53 database(http://p53.free.fr/Database
p53_database.html)
 Literature
 Institutionmutationcollections

23. Protein
stability Functional interactions
Protein damage
…KKRHCSGWL…
Unspecific cellular
Y interactions

24. Conceptual context: impact of mutations
on protein structure/function
 Empirical rules from site-directed
mutagenesis (‟80s, „90s):
◦ break disulphide bridges, burial of
charged residues, hydrogen bond loss,
disturb protein-protein interface, etc
◦ protein structure destabilization is
associated to function loss
 Evolutionary conservation is linked to
biological function

25. Mutation properties
 Sequence-based: V, , Blosum62
elements, etc
 Structure-based: relate to mutation
location: accessibility, contact number
and type, etc
 Evolutionary-based properties:
◦ wild-type (wt) conservation degree
◦ mutant rarity
◦ sequence variability at the mutation locus
(entropy)

26. Multiple alignments
 Low similarity, only two sequences:
AVTTGLNMWTTAKRPGMDDFYTILLPGLMNCI
GLFTAIDMHFFGRKPACEEYFTLVVDGLCNCI
 Low similarity, multiple sequences:
GIFTDIDMHFYVKKPGLDEFFTLVLRTLCMAA
ALTTGIDMWTTAKRPDMDDYYTIIIPGLMNCI
AVTTGLNMWTTAKRPGMDDFYTILLPGLMNCI
GVTTGLNMYFTARRPGLDEFYTLVLRTLCMCL
GIFTDIDMHFYVKKPGLDEFFTLVLRTLCMAA
AVTTGLNMWTTAKRPGMDDFYTILLPGLMNCI
GLFTALNMHFFGRKPACEEYFTLVVDGLCNCI

27. MSA: thetechnicalside
 Forverydivergentproteinsgood MSA
are veryhard to obtain
 Protocol to buildalignments:
◦ RecoverfamilymemberswithPsiBlast (E-
value:0.001; seq.id.>40%) UniRef100
◦ Align with MUSCLE
 Conservation may be misleading:
◦ proteinfunctionishighlyrelevantfor living
beings. E.g. histones. OK !
◦ databasebias. E.g.
onlyhominidaesequences are available.
PROBLEM ?!

29. Ourpredictor
 7 properties: sequence-based ( V, ,
Blosum62), structure (relativeaccessibility), MSA-
based (entropy, pssm(wt), pssm(mt))
 Neural networks (Wekapackage)
◦ Multilayerpercetron (1 hiddenlayer-4
units)
◦ No hiddenlayer
 Training: 2-fold cross-
validationscheme (25 replicas to

30. Performance measure: Matthews
correlationcoefficient
 MCC=(tp.tn-fp.fn)/[(tp+fp)(tp+fn)(tn+fp)(tn+fn)]1/2
 -1≤ MCC ≤1
◦ 0: predictivepower similar to random
◦ 1:perfectpredictionpower
◦ -1: badprediction, smallsamples, theproblemcannot be
solved?

31. ult

32. Pathogenicity prediction in Fabry
disease
 Mutationdataset: 313 pathological and
59 neutral mutations
 Discriminantpowerofparameters
 Performance

33. Aminoacid volume

34. Residueconservation

35. Performance

36. ROC curves

37. Performance (Successrate)

38. Performance (MCC)

39. GENE-SPECIFIC
PREDICTORS
-galactosidase PREDICTOR
MYH7 PREDICTOR
……… PREDICTOR

40. MYH7 (Beta-cardiac myosin heavy
chain)
 Largestructuralprotein (1390aa)
 Mutations cause familial
hypertrophiccardiomyopathy 1
 MutationdatasetobtainedfromUniProt,
OMIM andCardioGenomics
◦ 74 disease-causingmutations
◦ 45 neutral mutations (MSA)

41. Performance (Successrate)

42. Performance (MCC)

43. Gene-specific performances
Qtot= (tp+tn)/(tp+tn+fp+fn) Sensitivity= tp/(tp+fn) Specificity= tn/(tn+fp)
(neutral) (pathological)

44. Futuredirections:
pathogenicityprediction/analysis
 Extend to more genes:
◦ Enoughmutation data
◦ Notenoughmutation data
 Can
wepredictotherdiseasephenotypes?
◦ First tests suggest a similar
approachcouldworkforseverity

45. Summary
 Thereisroomforimprovement in
mutationannotationtools
 We are developping a new, gene-
basedtoolthatimprovespresentmethod
s
 Ourmethodwillworkforlarge-
scalescoringprojects (exome) andfor
single-mutationanalyses

46. WORKING TOGETHER

48. Towards a uniquemutationdamagereport
 Standardizethedescription/reportingof
mutationimpact:
◦ Sequence-level
◦ Structure-level
◦ MSA
◦ Miscellaneousinformation
 Communityeffort

49. TRANSLATIONAL BIOINFORMATICS
IN NEUROSCIENCES GROUP
•Neurovascular Disease, Neurosciences
•Joan Montaner
•Israel Fernández-Cadenas
•Nanomed.lysos.storage diseas., CIBBIM, Nanomedicine
•M.Carmen Domínguez
•Immunology, Respiratory & Systemic Diseases:
•Ricardo Pujol
•Mónica Martínez
•Roger Colobran
•Neuromusc.& Mitoch.Pathol, Neurosciences:
•Elena García
•Tomás Pinós
•Cancer and Iron group, IMPPC:
•Mayka Sánchez
•Ricky Joshi
•Biomedicine & Translat. & Pediatrics Oncol., Oncology
•Jaume Reventos
•Eva Colás
•Andreas Doll
•Marina Rigau
•Marta García