w Kin Mut An integrated tool for the

w. Kin. Mut An integrated tool for the analysis and interpretation of mutations in human protein kinases 1 Spanish José MG Izarzugaza National Cancer Research Center (CNIO) for Biological Sequence Analysis (CBS) 2 Center

Protein Kinases: Metabolic switches of the cell PK Functions Cell Cycle Regulation Signal Transduction Angiogenesis Immune Evasion Proliferation Anti-apoptosis Metastasis … Disease / Cancer 2 José MG Izarzugaza - txema@cbs. dtu. dk

Kinases and Cancer: Some examples ~30, 000 articles refer to ‘kinase AND mutation AND cancer’ in Pub. Med. José MG Izarzugaza - txema@cbs. dtu. dk 3

w. Kin. Mut: Interpreting Kinase Mutations 4 http: //wkinmut. bioinfo. cnio. es José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Interpreting Kinase Mutations 5 José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Gene & Protein info 6 José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: PFAM Domain Information 7 José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Mutations onto structures 8 (Izarzugaza et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Annotation databases Mutations occurring at the same position, different Aa allowed Uniprot SAAPdb Kin. Mut. Base COSMIC General information, experimental characterization Structural consequences of mutations Kinase mutations and disease Somatic Mutations in Cancer José MG Izarzugaza - txema@cbs. dtu. dk 9

w. Kin. Mut: Assessing the Pathogenicity 10 José MG Izarzugaza - txema@cbs. dtu. dk

Methods to predict the pathogenicity of Muts. SIFT SNAP SNPs&GO Polyphen-2 PMUT Mutation. Assessor Torkamani 11 José MG Izarzugaza - txema@cbs. dtu. dk

Kin. Mut: A Kinase-Specific Predictor Mutation selection criteria - Source db Uniprot - Human protein kinases - Experimentally classified - Non-synonymous, non-truncating. Dataset Description - Disease set: 865 muts, 65 kinases - Neutral set: 2627 muts, 447 kinases - Best dataset according to Care et al. , 2007 - Unevenly distributed in kinase groups. 12 (Izarzugaza et al. , BMC Genomics 2012) José MG Izarzugaza - txema@cbs. dtu. dk

Features: General & Kinase-Specific - Mutations vectors of features - General + Kinase-specific features Gene Level - Membership to Kin. Base groups - “Propensity to disease” of GO terms Domain Level - Mutation within PFAM domains Residue Level - Amino acid types Hydrophobicity changes Sequence conservation (SIFT) Functional annotations (SW, Fire. DB) Phosphorylation propensity Specificity Determining Positions (SDPs) (Izarzugaza et al. , BMC Genomics 2012) 13 José MG Izarzugaza - txema@cbs. dtu. dk

Machine Learning: SVM Soft Margin (C) Support Vector Machine (SVM) - Kernel (ϕ) Radial Basis Function - Optimized Parameters: Soft margin (C=8) Radius (ϒ=6 x 10 -4) - 10 K-fold cross-validation Gaussian Parameter (γ) ϒ C 14 (Izarzugaza et al. , BMC Genomics 2012) José MG Izarzugaza - txema@cbs. dtu. dk

Kin. Mut: Kinase-Specific Prioritization of Muts. Method Scope Accuracy Precision Recall MCC Kin. Mut Kinase 83. 3% 60. 0% 75. 2% 0. 6 SNP&GO Kinase (re-run with our dataset) 82. 3% 62. 8% 77. 4% 0. 6 Torkamani Kinase (from paper) 77. 0% - - 0. 5 Mutation. Assessor Kinase (re-run with our dataset) 53. 8% 41. 6% 95. 6% 0. 5 SNAP Kinase (re-run with our dataset) 49. 4% 34% 93. 1% 0. 4 SIFT Kinase (re-run with our dataset) 77. 6% 37. 8% 27. 9% 0. 2 Method Scope Accuracy Precision Recall MCC SNP&GO Genome-wide 82. 0% 83. 0% 78. 0% 0. 7 Mutation. Assessor Genome-wide 79. 0% - - - SNAP Genome-wide 78. 2% 76. 7% 80. 2% - SIFT Genome-wide 68. 3% 66. 1% 56. 5% 0. 3 (Izarzugaza et al. , BMC Genomics 2012) José MG Izarzugaza - txema@cbs. dtu. dk 15

Top most informative features 1. 2. 3. 4. 5. 6. 7. 8. 9. Disease propensity of GO terms (log odds ratio) Specificity determining positions Change in Hydrophobicity Amino acid type PFAM domains (Any, Pkinase_tyr, Fn…) SIFT Protein Kinase group (TK, CAMK, CK 1, TKL…) Uniprot annotations Phosphorylation propensity Kinbase groups (bold: Kinase-specific features) 16 Specificity determining positions (Izarzugaza et al. , BMC Genomics 2012) José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Assessing the Pathogenicity 17 José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Literature searches with SNP 2 L: - Automatic literature mining. - Kinase mutation mentions. - Mapping to protein sequences. - False positive filtering. 18 (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk

SNP 2 L: A Text Mining Pipeline 19 (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk

SNP 2 L: Simplified Pipeline Abstracts / Fulltext Mutation Detection … V 600 E B-Raf, which confers increased kinase activity, … Protein Detection … V 600 E B-Raf, which confers increased kinase activity, … Mutation Protein V 600 E B-Raf Filtering: Sequence Match 583 GDFGLAT V KSRWSG 606 600 20 Abstracts 643 different muts in 128 kinases Full Texts 6970 different muts in 325 kinases (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk

SNP 2 L: Manual Validation (100 abstracts) Ambiguous cases (2%) Wrong Mutation Detection (3%) Wrong Mutation to Protein Assignment (23%) Already in DBs (23%) Orthologues (8%) Correct, Manual Validation (41%) ~72% Mutation extractions proved correct… … including ~49% that were not already in databases. (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk 21

SNP 2 L: Structural distribution of mentions - Literature mentions distributed all over the PK domain. - Higher frequency associated to functional hot-spots. (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk 22

SNP 2 L: From Mutations to Literature Kinase Mutations SNP 2 L Evidence Phenotype Biochemical mechanism Organism/Tissue Experimental conditions … 23 (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Literature searches with SNP 2 L: - Automatic literature mining. - Kinase mutation mentions. - Mapping to protein sequences. - False positive filtering. 24 (Krallinger et al. , BMC Bioinformatics 2009) José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Interactions from Literature (i. Hop) 25 (Hoffman & Valencia, Nature Methods 2004) José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut: Integration of mutation information Gene/Protein Features Info in other Databases i. Hop Interactions Domain Features EGFR Gly 719 Ala : Lung cancer, somatic mutation Located in ATP binding pocket Confers resistance to gefitinib SNP 2 L Literature Residue Features 26 http: //wkinmut. bioinfo. cnio. es

Personalized (Stratified) Medicine “Here is my sequence” Sequencing and Genome Analysis José MG Izarzugaza - txema@cbs. dtu. dk 27

Simplified Personalized Med. Pipeline Blood (normal tissue) Cancer cells Educated Decision w. Kin. Mut Treatment Mutations Information Integration Personalized Medicine Cancer Patient Clinician José MG Izarzugaza - txema@cbs. dtu. dk 28

Integrate Information, help the clinician decide 29 (Courtesy of M. Vazquez) José MG Izarzugaza - txema@cbs. dtu. dk

w. Kin. Mut as part of our Pers. Med. Pipeline Endocrine Cancer Pancreas Cancer SIFT Polyphen-2 Mutation. Assessor Kin. Mut (kinases)

Acknowledgements Pathogenicity Prediction & Personalized Medicine (Izarzugaza et al. , BMC Genomics 2012) Alfonso Valencia Miguel Vazquez Angela del Pozo Victor de la Torre Text Mining (Izarzugaza et al. , Frontiers 2012) (Krallinger et al. , BMC Bioinformatics 2009) Martin Krallinger Mutation Mapping & Distribution Analysis (Izarzugaza et al. , Proteins 2009) (Izarzugaza et al. , BMC Bioinformatics 2011) Gonzalo Lopez Antonio Rausell Christine Orengo’s Group (UCL) Ollie Redfern Corin Yeats Benôit Desailly Andrew Martin’s Group (UCL) Anja Baresic Lisa Hopcroft 31 José MG Izarzugaza - txema@cbs. dtu. dk