Introduction to translational and clinical bioinformatics

Transcription

1 Introduction to translational and clinical bioinformatics Connecting complex molecular information to clinically relevant decisions using molecular profiles Constantin F. Aliferis M.D., Ph.D., FACMI Director, NYU Center for Health Informatics and Bioinformatics Informatics Director, NYU Clinical and Translational Science Institute Alexander Statnikov Ph.D. Director, Computational Causal Discovery laboratory Assistant Professor, NYU Center for Health Informatics and Bioinformatics, General Internal Medicine Director, Molecular Signatures Laboratory, Associate Professor, Department of Pathology, Adjunct Associate Professor in Biostatistics and Biomedical Informatics, Vanderbilt University 1

2 Goals Understand spectrum of Bioinformatics and Medical informatics activities Understand basic concepts of clinical/translational Bioinformatics Understand basic concepts of molecular profiling Introduction to high-throughput assays enabling molecular profiling Introduction to computational data analytics/bioinformatics enabling molecular profiling Understand analytic challenges and pitfalls/interpretation issues Discuss case study of profiles used to diagnose/treat patients Perform hands-on development of a molecular profile, finding novel biomarkers and testing profile/markers accuracy Discussion supported by general literature and heavily grounded on: - NYUMC informatics experts/research projects/grants/papers/entities/software systems - Commercially availiable modalities & assays 2

3 Overview Session #1: Basic Concepts Session #2: High-throughput assay technologies Session #3: Computational data analytics Session #4: Case study / practical applications Session #5: Hands-on computer lab exercise 3

4 Session #1: Basic Concepts Understand spectrum of Bioinformatics and Medical informatics activities - NYUMC informatics Understand basic concepts of clinical/translational Bioinformatics Understand basic concepts of molecular profiling ALSO: - s/names/interests - adjustments to plan 4

5 NYU Center for health Informatics & Bioinformatics: Broad Plan Health Informatics Infrastructure & Integrative Methods/Activities Bioinformatics Educational informatics Evidence based medicine, and Information retrieval Informatics Library Collaborative Data Integration & Mining: -Data warehouse & interfacing with EMR - Omics LIMS - Genomic EMR - Biospecimen management -research protocol database systems and management team -Data mining service -Data Mining software CTSI High Performance Computing Facility Research labs Kluger Molecular Signatures EBM, IR & Scientometrics Computational Causal Discovery MS/PhD (& Post-doc Fellowship) Program Continuing Education Workshops & tutorials Paper digest Research Colloquium Invited Speakers Integrate/Focus Existing Informatics and Increase Collaborations BPIC (best Practices Integrative Consultation Core/Service Literature Synthesis & Benchmarking studies Method-problem matchmaking Design and execution of studies Cancer Center Genetics- Genomics COEs Multi-modal & Integrative studies Proteomics Informatics Microarray Informatics: i. Upstream ii. Differential expression, iii. Pathway inference iv. Molecular profiles Next-gen sequencing informatics : Upstream analyses i. Chi-seq ii. RNA seq iii. Epigenetics iv. Microbiomics v. micro RNA studies vi. CNV & splice variation studies vii. Digital RNA viii. Denovo sequencing & resequencing Downstream analyses 5

6 Current Capabilities: Areas Health Informatics Infrastructure & Integrative Methods/Activities Bioinformatics Educational informatics Evidence based medicine, and Information retrieval Informatics Library Collaborative Data Integration & Mining: -Data warehouse & interfacing with EMR - Omics LIMS - Genomic EMR - Biospecimen management -research protocol database systems and management team -Data mining service -Data Mining software CTSI High Performance Computing Facility Research labs Kluger Molecular Signatures EBM, IR & Scientometrics Computational Causal Discovery MS/PhD (& Post-doc Fellowship) Program Continuing Education Workshops & tutorials Paper digest Research Colloquium Invited Speakers Integrate/Focus Existing Informatics and Increase Collaborations BPIC (best Practices Integrative Consultation Core/Service Literature Synthesis & Benchmarking studies Method-problem matchmaking Design and execution of studies Cancer Center Genetics- Genomics COEs Multi-modal & Integrative studies Proteomics Informatics Microarray Informatics: i. Upstream ii. Differential expression, iii. Pathway inference iv. Molecular profiles Next-gen sequencing informatics : Upstream analyses i. Chi-seq ii. RNA seq iii. Epigenetics iv. Microbiomics v. micro RNA studies vi. CNV & splice variation studies vii. Digital RNA viii. Denovo sequencing & resequencing Downstream analyses 6

7 Current & Future capabilities Health Informatics Educational informatics Evidence based medicine, and Information retrieval Informatics Content management, medical simulations Filter Medline according to content and quality Filter Web for health advice quality Predict future citations of articles Classify individual citations as instrumental or not Identify special types of articles Construct citation histories & Analyze impact of articles Integrate and manage queries and related content Combine and optimize knowledge source searches New find a researcher Find a collaborator Library Collaborative Apply, evaluate, refine next-gen IR methods Data Integration & Mining: -Data warehouse & interfacing with EMR - Omics LIMS - Genomic EMR - Biospecimen management -research protocol database systems and management team -Data mining service -Data Mining software -Data warehouse needs; software acquisition; implementation - OMICS LIMS needs capture; vendor product assessment; funds; sofwtare purchase and implementation; integration with billing and EMR -Biospecimen management -Research protocol database system (evelos) -Data base management team -Data mining service -Data mining engine: faculty; funds; prototype; implementation; evaluation 7

8 Current & Future capabilities Infrastructure & Integrative Methods/Activities CTSI High Performance Computing Facility Research labs Kluger Molecular Signatures EBM, IR & Scientometrics Computational Causal Discovery MS/PhD (& Post-doc Fellowship) Program (supported by rest of objectives) Sequencing server; hectar1; hectar2; Funds; needs; grants; personnel post; specs; room/networking/access; Personnel hires; hw install; licenses; BP; launch Kluger TF /Regulation studies; high-throughput outcome prediction, specialized clustering methods Molecular Signatures development of molecular signatures for diagnosis outcome prediction and personalized medicine, discovery of diagnostic/imaging biomarkers and putative drug targets, deployment of signatures, automated software, new methods EBM, IR & Scientometrics development and evaluation of next-gen IR and scientometric models and studies Computational Causal Discovery discovery of pathways; studies of causal validity of bioinformatics discovery methods, multiplicity studies, automated software, active learning/experiment number minimization Formal Training in Biomedical Informatics at pre and post-doctoral levels Continuing Education Workshops & tutorials Paper digest Research Colloquium Invited Speakers Workshops & tutorials Paper digest Research Colloquium Invited Speakers Continuing Education Integrate/Focus Existing Informatics and Increase Collaborations Faculty and Staff career development; Informatics Affiliates; Working Collaborations with Courant, Polytechnic, NYC Informatics and other non-nyumc entities 8

9 Current & Future capabilities Bioinformatics BPIC (best Practices Integrative Consultation Core/Service Literature Synthesis & Benchmarking studies Method-problem matchmaking Design and execution of studies Study publication assistance Area-specific (Disease, Assay) Informatics Genetics-Genomics COEs Cancer Center Microarray Informatics: Experiment design, assay execution, differential expression, pathway mapping, pathway-specific testing (GSEA/GSA), de novo pathway discovery, phylogeny, clustering, hybrid experimental/observational designs; SNP arrays; ChIPon-ChIP analyses, acgh, tiled arrays, etc Sequencing Informatics: Chip-Seq analysis, digital gene expression, de novo sequence assembly & reassembly, CNV analysis, epigenomic studies, microbiomics Proteomics Informatics: platform-specific pre-processing, differential abundance, peptide-protein mapping, protein identification, de novo protein interaction network inference, protein modification and structure studies, Multi-modal Integrative and Higher-level Informatics: Molecular Signatures & linking high-dimensional data to phenotype development of molecular signatures for diagnosis, outcome prediction and personalized medicine; in silico signature scanning, in silico signature equivalence, discovery of diagnostic/imaging biomarkers and putative drug targets, deployment of signatures, automated software, novel methods Mechanistic /causative studies discovery of pathways; multiplicity studies, TS/DBN designs, automated software, active learning/experiment number minimization Integrating clinical lab, text, imaging and high throughput data in CTs/prospective studies or exploratory retrospective ones 9

10 Summary Contacts (Until Centralized Consultation Service is Launched) Management of Clinical and protocol data Educational Informatics Next-Gen Information Retrieval Informatics for Data Mining Data Integration & Warehousing High Performance Computing Best Practices in Bioinformatics Sequencing Informatics Microarray Informatics Cancer Informatics Proteomics Informatics General Tools Specialized applications (Genetics, Regulation, Pathways ) Molecular Signatures development, biomarker discovery, Multi-modal and Integrative studies James Robinson Mark Triola Lawrence Fu, Constantin Aliferis, TBD Alexander Statnikov, Constantin Aliferis John Chelico, Ross Smith, Constantin Aliferis Constantin Aliferis, Ross Smith Constantin Aliferis, Alexander Statnikov Upstream: Stuart Brown, Alexander Alekseyenko, Yuval Kluger, Jinhua Wang, TBD, TBD Downstream: Alexander Alekseyenko, Yuval Kluger, Jinhua Wang, Alexander Statnikov, Constantin Aliferis Jiri Zafadil, Yuval Kluger, Jinhua Wang, Constantin Aliferis, Alexander Statnikov Yuval Kluger, Jinhua Wang, Stuart Brown, Jiri Zafadil, Constantin Aliferis Stuart Brown, Jinhua Wang, Constantin Aliferis, Alexander Statnikov, TBD Stuart Brown Stuart Brown, Yuval Kluger, Alexander Statnikov, Constantin Aliferis Constantin Aliferis, Alexander Statnikov, Yuval Kluger 10

11 Molecular Signatures Definition = computational or mathematical models that link highdimensional molecular information to phenotype of interest 11

12 Molecular Signatures Gene markers New drug targets 12

13 Molecular Signatures: Main Uses 1. Direct benefits: Models of disease phenotype/clinical outcome & estimation of the model performance Diagnosis Prognosis, long-term disease management Personalized treatment (drug selection, titration) ( predictive models) 2. Ancillary benefits 1: Biomarkers for diagnosis, or outcome prediction Make the above tasks resource efficient, and easy to use in clinical practice Helps next-generation molecular imaging Leads for potential new drug candidates 3. Ancillary benefits 2: Discovery of structure & mechanisms (regulatory/interaction networks, pathways, sub-types) Leads for potential new drug candidates 13

14 Molecular Signatures The FDA calls them in vitro diagnostic multivariate index assays 1. Class II Special Controls Guidance Document: Gene Expression Profiling Test System for Breast Cancer Prognosis : - addresses device classification 2. The Critical Path to New Medical Products : - identifies pharmacogenomics as crucial to advancing medical product development and personalized medicine. 3. Draft Guidance on Pharmacogenetic Tests and Genetic Tests for Heritable Markers & Guidance for Industry: Pharmacogenomic Data Submissions - identifies 3 main goals (dose, ADEs, responders), - define IVDMIA, - encourages fault-free sharing of pharmacogenomic data, - separates probable from valid biomarkers, - focuses on genomics (and not other omics), 14

15 Less Conventional Uses of Molecular Signatures Increased Clinical Trial sample efficiency, and decreased costs or both, using placebo responder signatures ; In silico signature-based candidate drug screening; Drug resurrection Establishing existence of biological signal in very small sample situations where univariate signals are too weak; Assess importance of markers and of mechanisms involving those Choosing the right animal model? 15

16 Recent molecular mignatures available for patient care Agendia Clarient Prediction Sciences LabCorp OvaSure University Genomics Genomic Health Veridex BioTheranostics Applied Genomics Power3 Correlogic Systems 16

17 Molecular signatures in the market (examples) Company Product Disease Purpose Agendia MammaPrint Risk assessment for the recurrence of distant metastasis in a breast Breast cancer cancer patient. Agendia TargetPrint Quantitative determination of the expression level of estrogen receptor, Breast cancer progesteron receptor and HER2 genes. This product is supplemental to MammaPrint. Agendia CupPrint Cancer Determination of the origin of the primary tumor. University Genomics Breast Bioclassifier Breast cancer Classification of ER-positive and ER-negative breast cancers into expression-based subtypes that more accurately predict patient outcome. Clarient Clarient Prediction Sciences Insight Dx Breast Cancer Profile Prostate Gene Expression Profile RapidResponse c-fn Test Breast cancer Prediction of disease recurrence risk. Prostate cancer Stroke Genomic Health OncotypeDx Breast cancer Diagnosis of grade 3 or higher prostate cancer. Identification of the patients that are safe to receive tpa and those at high risk for HT, to help guide the physician s treatment decision. Individualized prediction of chemotherapy benefit and 10-year distant recurrence to inform adjuvant treatment decisions in certain women with early-stage breast cancer. biotheranostics CancerTYPE ID Cancer Classification of 39 types of cancer. Risk assessment and identification of patients likely to benefit from biotheranostics Breast Cancer Index Breast cancer endocrine therapy, and whose tumors are likely to be sensitive or resistant to chemotherapy. Applied Genomics MammaStrat Breast cander Risk assessment of cancer recurrence. Applied Genomics Applied Genomics PulmoType PulmoStrat Lung cancer Lung cancer Classification of non-small cell lung cancer into adenocarcinoma versus squamous cell carcinoma subtypes. Assessment of an individual's risk of lung cancer recurrence following surgery for helping with adjuvant therapy decisions. Correlogic OvaCheck Ovarian cancer Early detection of epithelial ovarian cancer. Assessment of the presence of early stage ovarian cancer in high-risk LabCorp OvaSure Ovarian cancer women. Veridex GeneSearch BLN Assay Breast cancer Determination of whether breast cancer has spread to the lymph nodes. Power3 BC-SeraPro Breast cancer Differentiation between breast cancer patients and control subjects. 17

18 MammaPrint Developed by Agendia ( 70-gene signature to stratify women with breast cancer that hasn t spread into low risk and high risk for recurrence of the disease Independently validated in >1,000 patients So far performed 12,000 tests Cost of the test is $3,200 In February, 2007 the FDA cleared the MammaPrint test for marketing in the U.S. for node negative women under 61 years of age with tumors of less than 5 cm. TIME Magazine s 2007 medical invention of the year. 18

19 CupPrint Developed by Agendia ( ~500-gene (~1900 probes) signature to identify primary site of 49 different types of carcinomas as well as other types of cancer such as sarcoma and melanoma. Several independent validation studies 19

20 ColoPrint In development & validation by Agendia ( Multi-gene expression signature to determine the risk for recurrence in colorectal cancer patients Planning to seek FDA approval References:

21 Oncotype DX Development synopsis Main reference: Paik S, Shak S, Tang G, et al. A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. N Engl J Med. 2004; 351(27): Developed by Genomic Health ( ) 21-gene signature to predict whether a woman with localized, ER+ breast cancer is at risk of relapse Independently validated in >1,000 patients So far performed 55,000 tests Cost of the test is $3,650 Reimbursement. Information about reimbursement for molecular signatures from Aetna: Oncotype DX did not undergo FDA review. Here is an article that mentions FDA review of Oncotype DX (slightly outdated): The following paper shows the health benefits and cost-effectiveness benefits of using Oncotype DX: 21

22 CancerType ID Developed by AviaraDX ( 92-gene signature to classify 39 tumor types Signature developed by GA/KNN Compressed version of CupPrint 22

23 Breast Cancer Index Developed by AviaraDX ( Uses 7 genes (combines 5-gene MGI signature and 2-gene H/I signature) Stratifies breast cancer patients into groups with low or high risk of cancer recurrence and good or poor response to endocrine therapy. Validated in thousands of patients (treated & untreated) 23

24 GeneSearch Breast Lymph Node (BLN) Assay Developed by Veridex ( a Johnson & Johnson company Test to detect if breast cancer has spread to the lymph nodes The GeneSearch BLN uses real-time reverse transcriptase-polymerase chain reaction (RT-PCR) to detect ammoglobin (MG) and cytokeratin 19 (CK 19) in lymph nodes. FDA approved Featured in TIME s 2007 Top 10 Medical Breakthroughs list 24

25 MammoStrat Developed by Applied Genomics ( The test is based on 5 biomarkers. The test is used to classify individual patients as having an AGI-defined high-, moderate-, or low-risk of breast cancer recurrence following surgical removal of their primary tumor and treatment with tamoxifen alone. Independently validated in >1000 patients 25

26 NuroPro Developed by Power3 ( Early detection of neurodegenerative diseases: Alzheimer s disease, ALS (Lou Gehrig s disease), and Parkinson s disease. Validation study in progress. Based on 59 proteins. 26

27 BC-SeraPro Developed by Power3 ( Test for diagnosis of breast cancer (breast cancer case vs. control). Validation study in progress. Based on 22 proteins. Uses linear discriminant analysis; outputs a probability score. 27

28 Key ingredients for developing a molecular signature Well-defined clinical problem & access to patients Computational & Biostatistical Analysis Molecular Signature High-throughput assays 28

29 Challenges in Computational Analysis of omics data for development of molecular signatures Relatively easy to develop a predictive model + even easier to believe that a model is good when it is not false sense of security Several problems exist: some theoretical and some practical Omics data has many special characteristics and is tricky to analyze! 29

30 OvaCheck Developed by Correlogic ( Blood test for the early detection of epithelial ovarian cancer Failed to obtain FDA approval Looks for subtle changes in patterns among the tens of thousands of proteins, protein fragments and metabolites in the blood Signature developed by genetic algorithm Significant artifacts in data collection & analysis questioned validity of the signature: - Results are not reproducible - Data collected differently for different groups of patients 30

31 Problem with OvaCheck Data Set 1 (Top), Data Set 2 (Bottom) Cancer A Normal B Other Cancer C D Figure from Baggerly et al (Bioinformatics, 2004) Normal E Other F Clock Tick 31

32 Molecular Signatures Gene markers New drug targets 32

33 Brief History of main omics technology: gene expression microarrays 1988: Edwin Southern files UK patent applications for in situ synthesized, oligonucleotide microarrays 1991: Stephen Fodor and colleagues publish photolithographic array fabrication method 1992: Undeterred by NIH naysayers, Patrick Brown develops spotted arrays 1993: Affymax begets Affymetrix 1995: Mark Schena publishes first use of microarrays for gene expression analysis Edwin Southern founds Oxford Gene Technologies 1996: First human gene expression microarray study published Affymetrix releases its first catalog GeneChip microarray, for HIV, in April 1997: Stanford researchers publish the first whole-genome microarray study, of yeast 33

34 Brief History of main omics technology: gene expression microarrays (The scientist 2005) 1998: Brown's lab develops CLUSTER, a statistical tool for microarray data analysis; red and green "thermal plots" start popping up everywhere 1999: Todd Golub and colleagues use microarrays to classify cancers, sparking widespread interest in clinical applications 2000: Affymetrix spins off Perlegen, to sequence multiple human genomes and identify genetic variation using arrays 2001: The Microarray Gene Expression Data Society develops MIAME standard for the collection and reporting of microarray data 2003: Joseph DeRisi uses a microarray to identify the SARS virus Affymetrix, Applied Biosystems, and Agilent Technologies individually array human genome on a single chip 2004: Roche releases Amplichip CYP450, the first FDA-approved microarray for diagnostic purposes 34

35 An early kind of analysis: learning disease sub-types by clustering patient profiles p53 Rb 35

36 Clustering: seeking natural groupings & hoping that they will be useful p53 Rb 36

37 E.g., for treatment Respond to treatment Tx1 p53 Do not Respond to treatment Tx1 Rb 37

38 E.g., for diagnosis Adenocarcinoma p53 Squamous carcinoma Rb 38

39 Another use of clustering Cluster genes (instead of patients): Genes that cluster together may belong to the same pathways Genes that cluster apart may be unrelated 39

40 Unfortunately clustering is a non-specific method and falls into the one-solution fits all trap when used for prediction Do not Respond to treatment Tx2 p53 Respond to treatment Tx2 Rb 40

41 Clustering is also non-specific when used to discover pathway membership, regulatory control, or other causation-oriented relationships G1 Ph G2 G3 It is entirely possible in this simple illustrative counter-example for G3 (a causally unrelated gene to the phenotype) to be more strongly associated and thus cluster with the phenotype (or its surrogate genes) more strongly than the true oncogenic genes G1, G2 41

42 Two improved classes of methods Supervised learning predictive signatures and markers Regulatory network reverse engineering pathways 42

43 Supervised learning : use the known phenotypes (a.k.a labels) in training data to build signatures or find markers highly specific for that phenotype A B C D E TRAIN INSTANCES INDUCTIVE ALGORITHM Classifier OR Regression Model APPLICATION INSTANCES A 1, B 1, C 1, D 1, E 1 A 2, B 2, C 2, D 2, E 2 A n, B n, C n, D n, E n CLASSIFICATION PERFORMANCE 43

44 Regulatory network reverse engineering B A C TRAIN INSTANCES INDUCTIVE ALGORITHM B A C E D E D A 1, B 1, C 1, D 1, E 1 A 2, B 2, C 2, D 2, E 2 A n, B n, C n, D n, E n PERFORMANCE 44

45 Supervised learning: a geometrical interpretation p53 Cancer patients New case, classified as cancer P1 SVM classifier? P2 + P3 + P P4? New case, classified as normal + Normals Rb 45

46 In 2-D looks good but what happens in: 10,000-50,000 (regular gene expression microarrays, acgh, and early SNP arrays) >500,000 (tiled microarrays, new SNP arrays) 10, ,000 (regular MS proteomics) >10, 000, 000 (LC-MS proteomics) This is the curse of dimensionality problem 46

47 High-dimensionality (especially with Some methods do not run at all (classical regression) Some methods give bad results (KNN, Decision trees) Very slow analysis Very expensive/cumbersome clinical application Tends to overfit small samples) causes: 47

48 Two (very real and very unpleasant) problems: Over-fitting & Under-fitting Over-fitting ( a model to your data)= building a model that is good in original data but fails to generalize well to fresh data Under-fitting ( a model to your data)= building a model that is poor in both original data and fresh data 48

49 Intuitive explanation of overfitting & underfitting Play the game: find rule to predict who are the instructors in any given class (use today s class to find a general rule) 49

50 Over/under-fitting are directly related to the complexity of the decision surface and how well the training data is fit Outcome of Interest Y This line is good! Training Data Future Data This line overfits! Predictor X 50

51 Over/under-fitting are directly related to the complexity of the decision surface and how well the training data is fit Outcome of Interest Y This line is good! Training Data Future Data This line underfits! Predictor X 51

52 Very Important Concept: Successful data analysis methods balance training data fit with complexity. Too complex signature (to fit training data well) overfitting (i.e., signature does not generalize) Too simplistic signature (to avoid overfitting) underfitting (will generalize but the fit to both the training and future data will be low and predictive performance small). 52

53 Part of the Solution: feature selection P A O B C D E T K H I J Q L M N 53

54 How well supervised learning works in practice? 54

55 Datasets Bhattacharjee2 - Lung cancer vs normals [GE/DX] Bhattacharjee2_I - Lung cancer vs normals on common genes between Bhattacharjee2 and Beer [GE/DX] Bhattacharjee3 - Adenocarcinoma vs Squamous [GE/DX] Bhattacharjee3_I - Adenocarcinoma vs Squamous on common genes between Bhattacharjee3 and Su [GE/DX] Savage - Mediastinal large B-cell lymphoma vs diffuse large B-cell lymphoma [GE/DX] Rosenwald4-3-year lymphoma survival [GE/CO] Rosenwald5-5-year lymphoma survival [GE/CO] Rosenwald6-7-year lymphoma survival [GE/CO] Adam - Prostate cancer vs benign prostate hyperplasia and normals [MS/DX] Yeoh - Classification between 6 types of leukemia [GE/DX-MC] Conrads - Ovarian cancer vs normals [MS/DX] Beer_I - Lung cancer vs normals (common genes with Bhattacharjee2) [GE/DX] Su_I - Adenocarcinoma vs squamous (common genes with Bhattacharjee3) [GE/DX Banez - Prostate cancer vs normals [MS/DX] 55

56 Methods: Gene and Peak Selection Algorithms ALL - No feature selection LARS - LARS HITON_PC - HITON_PC_W - HITON_PC+ wrapping phase HITON_MB - HITON_MB_W - HITON_MB + wrapping phase GA_KNN - GA/KNN RFE - RFE with validation of feature subset with optimized polynomial kernel RFE_Guyon - RFE with validation of feature subset with linear kernel (as in Guyon) RFE_POLY - RFE (with polynomial kernel) with validation of feature subset with polynomial optimized kernel RFE_POLY_Guyon - RFE (with polynomial kernel) with validation of feature subset with linear kernel (as in Guyon) SIMCA - SIMCA (Soft Independent Modeling of Class Analogy): PCA based method SIMCA_SVM - SIMCA (Soft Independent Modeling of Class Analogy): PCA based method with validation of feature subset by SVM WFCCM_CCR - Weighted Flexible Compound Covariate Method (WFCCM) applied as in Clinical Cancer Research paper by Yamagata (analysis of microarray data) WFCCM_Lancet - Weighted Flexible Compound Covariate Method (WFCCM) applied as in Lancet paper by Yanagisawa (analysis of mass-spectrometry data) UAF_KW - Univariate with Kruskal-Walis statistic UAF_BW - Univariate with ratio of genes between groups to within group sum of squares UAF_S2N - Univariate with signal-to-noise statistic 56

57 Classification Performance (average over all tasks/datasets) 57

58 How well gene selection works in practice? 58

59 ALL LARS HITONgp_PC HITONgp_MB HITONgp_PC_W HITONgp_MB_W GA_KNN RFE RFE_Guyon RFE_POLY RFE_POLY_Guyon SIMCA SIMCA_SVM WFCCM_CCR UAF_KW UAF_BW UAF_S2N Number of Selected Features (average over all tasks/datasets)

60 Number of Selected Features (zoom on most powerful methods) LARS HITONgp_PC HITONgp_MB HITONgp_PC_W HITONgp_MB_W GA_KNN RFE RFE_Guyon RFE_POLY RFE_POLY_Guyon 60

61 Number of Selected Features (average over all tasks/datasets) 61

62 Conclusions so far Special classifiers (with inherent complexity control) combined with feature selection & careful parameterization protocols overcome over-fitting & estimate future performance accurately. Caveats: analysis is typically complex and error prone. Need: (a) an experienced analyst on the team, or (b) a validated software system designed for nonexperts. 62

63 Software Causal Explorer Gems Fast-aims 63

64 Causal Explorer Matlab library of computational causal discovery and variable selection algorithms Introductory-level library to our causal algorithms (~3% of our algorithms) Discover the direct causal or probabilistic relations around a response variable of interest (e.g., disease is directly caused by and directly causes a set of variables/observed quantities). Discover the set of all direct causal or probabilistic relations among the variables. Discover the Markov blanket of a response variable of interest, i.e., the minimal subset of variables that contains all necessary information to optimally predict the response variable. Code emphasizes efficiency, scalability, and quality of discovery Requires relatively deep understanding of underlying theory and how the algorithms operate 64

65 Statistics of Registered Users 739 registered users in >50 countries. 402 (54%) users are affiliated with educational, governmental, and non-profit organizations 337 (46%) users are either from private or commercial sectors. Major commercial organizations that have registered users of Causal Explorer include: IBM Intel SAS Institute Texas Instruments Siemens GlaxoSmithKline Merck Microsoft 65

66 Statistics of Registered Users Major U.S. institutions that have registered users of Causal Explorer: Boston University Brandies University Carnegie Mellon University Case Western Reserve University Central Washington University College of William and Mary Cornell University Duke University Harvard University Illinois Institute of Technology Indiana University-Purdue University Indianapolis Johns Hopkins University Louisiana State University M. D. Anderson Cancer Center Massachusetts Institute of Technology Medical College of Wisconsin Michigan State University Naval Postgraduate School New York University Northeastern University Northwestern University Oregon State University Pennsylvania State University Princeton University Rutgers University Stanford University State University of New York Tufts University University of Arkansas University of California Berkley University of California Los Angeles University of California San Diego University of California Santa Cruz University of Cincinnati University of Colorado Denver University of Delaware University of Houston-Clear Lake University of Idaho University of Illinois at Chicago University of Illinois at Urbana- Champaign University of Kansas University of Maryland Baltimore County University of Massachusetts Amherst University of Michigan University of New Mexico University of Pennsylvania University of Pittsburgh University of Rochester University of Tennessee Chattanooga University of Texas at Austin University of Utah University of Virginia University of Washington University of Wisconsin- Madison University of Wisconsin- Milwaukee Vanderbilt University Virginia Tech Yale University 66

67 Other systems for supervised analysis of microarray data Name Version Developer Automatic model selection for classifier and gene selection methods ArrayMiner ClassMarker 5.2 Optimal Design, Belgium No Avadis Prophetic 3.3 Strand Genomics, USA No BRB ArrayTools 3.2 Beta National Cancer Institute, USA No University of Pittsburgh and University of cageda supervised (accessed 10/2004) Pittsburgh analysis No Medical Center, of USA microarray data, but Cleaver 1.0 (accessed 10/2004) GeneCluster Stanford University, USA GeneLinker Platinum 4.5 Predictive Patterns Software, Canada No GeneMaths XT 1.02 Applied Maths, Belgium No Broad Institute, Massachusetts Institute of GenePattern No algorithms Technology, or algorithms USA with unknown performance. Genesis Graz University of Technology, Austria No GeneSpring 7 Silicon Genetics, USA No GEPAS There exist many good software packages for Neither system Broad Institute, provides Massachusetts a Institute protocol of for data analysis No that Technology, USA precludes overfitting. A typical software either offers an overabundance of The software packages address needs only of experienced analysts. 1.1 (accessed 10/2004) National Center for Cancer Research (CNIO), Spain No Limited (for number of genes) MultiExperiment Viewer The Institute for Genomic Research, USA No PAM 1.21a Stanford University, USA Partek Predict 6.0 Partek, USA Limited (for a single parameter of the classifier) Limited (does not allow optimization of the choice of gene selection algorithms) Weka Explorer University of Waikato, New Zeland No 67

68 Purpose of GEMS Gene expression data and outcome variable Normal Cancer Cancer Normal Normal Cancer Cancer Cancer Normal Classification model Cross-validation performance estimate Optional: Gene names & IDs ring finger protein 1 tubulin, beta, 5 glucose-6-phosphate dehydrogenase glutathione S-transferase M5 carnitine acetyltransferase Rho GTPase activating protein 4 SMA3 mannose phosphate isomerase mitogen-activated protein kinase 3 leukotriene A4 hydrolase chromosome 21 open reading frame 1 dihydropyrimidinase-like 2 beta-2-microglobulin discs, large (Drosophila) homolog 4 (model generation & performance estimation mode) Reduced set of genes Rho GTPase activating protein 4 SMA3 mannose phosphate isomerase mitogen-activated protein kinase 3 Links to literature 68

69 Purpose of GEMS Gene expression data and unknown outcome variable????????? Model predictions Normal Cancer Cancer Normal Normal Cancer Cancer Cancer Normal Classification model Performance estimate (model application mode) 69

70 MC-SVM Methods Implemented in GEMS Cross-Validation Designs N-Fold CV LOOCV Normalization Techniques [a, b] (x MEAN(x)) / STD(x) Classifiers One-Versus-Rest One-Versus-One DAGSVM Method by WW Method by CS Gene Selection Methods S2N One-Versus-Rest S2N One-Versus-One Non-param. ANOVA x / STD(x) BW ratio x / MEAN(x) x / MEDIAN(x) x / NORM(x) Performance Metrics Accuracy HITON_MB HITON_PC x MEAN(x) RCI x MEDIAN(x) AUC ROC ABS(x) x + ABS(x) 70

71 Software Architecture of GEMS GEMS 2.0 Wizard-Like User Interface Computational Engine Estimate classification performance Generate a classification model and estimate its performance Generate a classification model Apply existing model to a new set of patients Report Generator I II X Cross-Validation Loop for Performance Est. N-Fold CV LOOCV I Cross-Validation Loop for Model Selection N-Fold CV I LOOCV II II Performance Computation Accuracy RCI AUC ROC Normalization Gene Selection S2N One-Versus-Rest S2N One-Versus-One Non-param. ANOVA BW ratio HITON_PC HITON_MB Classification by MC-SVM One-Versus-Rest One-Versus-One DAGSVM Method by WW Method by CS 71

72 GEMS 2.0: Wizard-Like Interface Task selection Dataset specification Cross-validation design Normalization Logging Performance metric Gene selection Classification Report generation Analysis execution 72

73 GEMS 2.0: Wizard-Like Interface Input microarray gene expression dataset File with gene names File with gene accession numbers Output model 73

74 Statistics of registered users 800 users in >50 countries 350 academic & non-profit users 450 private & commercial users 205 scientific citations of major paper that introduced GEMS Major commercial organizations that have registered users of Causal Explorer include: Eli Lilly Novartis IBM GE Genedata Nuvera Biosciences GenomicTree Cogenetics Pronota 74

75 FAST-AIMS FAST-AIMS is a system to support automatic development of high-quality classification models and biomarker discovery in mass spectrometry proteomics data Incorporates automated data analysis protocols of GEMS Deals with additional challenges of MS data analysis 75

76 System Workflow 76

77 Evaluation in multiple user study 77