A Novel Framework to Elucidate Core Classes in a Dataset

1. A Novel Frameworkto Elucidate Core Classes in a Dataset Daniel Soria d.soria@cs.nott.ac.uk G54DMT - Data Mining Techniques and Applications 22nd March 2011

2. Outline • Aims and motivations • Framework • Clustering and consensus • Supervised learning • Validation on two clinical data sets • Conclusions and questions

3. Aims and motivations • Develop an original framework with multiple steps to extract most representative classes from any dataset • Refine the phenotypic characterisation of breast cancer • Move the medical decision making process from a single technique approach to a multi-technique one • Guide clinicians in the choice of the favourite andmost powerful treatment to reach a personalised healthcare

4. Framework (1) no yes

5. Framework (2) yes no

6. Data pre-processing • Dealing with missing values • Homogeneous variables • Computedescriptive statistics

7. Clustering • Four different algorithms: • Hierarchical (HCA) • Fuzzy c-means (FCM) • K-means (KM) • Partitioning Around Medoids (PAM) • Methods run with the number of clusters varying from 2to20

8. Hierarchical method • Hierarchy of clusters • Represented with a tree (dendrogram) • Clusters obtained cutting the dendrogramat specific height

9. Hierarchical method (cont.) 6 clusters

10. FCM method • Minimisationof the o. f. • X = {x1,x2,...,xn}:ndata points • V = {v1,v2,…,vc}:ccluster centres • U=(μij)n*c:fuzzy partition matrix • μij:membership of xito vj • m:fuzziness index

11. FCM method (cont.)

12. KM method • Minimisationof the o. f. • ||xi-vj||:Euclidean distance betweenxiand vj • cj:data points in clusterj • vj

13. KM method (cont.)

14. PAM method • Search for krepresentative objects (medoids) among the observations • Minimum sum of dissimilarities • kclusters are constructed by assigning each observation to the nearest medoid

15. PAM method (cont.)

16. If n is unknown… • Validity indicescomputation • Defined considering the data dispersion within and between clusters • According to decision rules, the best number of clusters may be selected

17. Validity Indices

18. Characterisation & agreement • Visualisation techniques • Biplots • Boxplots • Indices for assessing agreement • Cohen’s kappa (κ) • Rand

19. Definition of classes • Consensus clustering: • Align labels to have similar clusters named in thesame way by different algorithms • Take into account points assigned to groups with the same label

20. Supervised learning (1) • Model-based classification for prediction of future cases • Aims • High quality prediction • Reduce number of variables (biomarkers) • Prefer ‘white-box’ prediction models

21. Supervised learning (2) • Different techniques: • C4.5 • Multi-Layer Perceptron Artificial Neural Network (MLP-ANN) • Naïve Bayes (NB) or Non-Parametric Bayesian Classifier (NPBC)

22. C4.5 classifier • Each attribute can be used to make a decision that splits the data into smaller subsets • Information gainthat results from choosing an attribute for splitting the data • Attribute with highest information gain is the one used to make the decision

23. Multi-Layer Perceptron • Feed-forward ANN • Nonlinear activation function used by each neuron • Layers of hidden nodes connected with every other node in the following layer • Learning carried out through back-propagation

24. Naïve Bayes classifier • Probabilistic classifier based on Bayes’ theorem • Good for multi-dimensional data • Common assumptions: • Independence of variables • Normality

25. NPBC: ratiobetweenareas (1) • Similar to Naïve Bayes • Useful for non-normal data • Based on ratio between areasunder the histogram • The closer a data point is to the median, the higher the probability to belong to that specific class 25

26. NPBC: ratiobetweenareas (2) x< m x > m datapointx median m median m datapointx ER ER Soria et al. (2011): A `Non-Parametric’ Version of the Naïve Bayes Classifier. To appear in Knowledge-Based Systems 26

27. Characterisation of classes • Biplots • Boxplots • Relation with clinical information • Survival analysis

28. Validation of the framework (1) • Set of markers involved in breast cancer cell cycle regulation • 347 patients and 4 markers • Survival and grade available • K-means, PAM and Fuzzy C-means used • Validity indices for best number of clusters

29. KM validity indices

30. PAM validity indices

31. Optimum number of clusters

32. FCM validity indices

33. Optimum number of clusters

34. Agreement and consensus • Kappa index high for 3 groups classification • 3 common classes found (8.9% not classified) • Intermediateexpression (class 1) • High expression (class 2) • Lowexpression (class 3)

35. Biplots of classes (1) °: Class 1 °: Class 2 °: Class 3

36. Biplots of classes (2) °: Class 1 °: Class 2 °: Class 3 °: N.C.

37. Boxplots of classes

38. Kaplan-Meier curves

39. Clinical information • High grade patients (poor prognosis) in classes 1 and 3 (worst survival) • Common classes group patients with similar outcome

40. C4.5 decision tree

41. Validation of the framework (2) • Patients entered into the NottinghamTenovus Primary Breast Carcinoma Series between 1986 and 1998 • 1076 cases informative for all 25 biological markers • Clinical information(grade, size, age, survival, follow-up, etc.) available

42. Previous work: 4 groups

43. Breast Cancer ER+ Luminal CKs+ ER- Basal CKs- ER- HER2+ Mixed Class (38.4%) • HER2 Class 6 (7.2%) Luminal Basal PgR+ PgR- p53+ p53- • p53 altered Class 4 (7.6%) p53 normal Class 5 (6.4%) HER3+ HER4+ HER3- HER4- HER3+ HER4+ Luminal A Class 1 (18.8%) • Luminal N Class 2 (14.2%) Luminal B Class 3 (7.4%) Consensus Clustering

44. Conclusions • Original framework for identification of core classes • Formed by different logical steps • Validation over novel data sets • 3 classes (Low, IntermediateandHigh) • High marker levels for better survival • Discover of novel cancer subtypes

45. Main references • L. Kaufman, P.J. Rousseeuw. Finding groups in data, Wiley series in probability and mathematical statistics, 1990. • A. Weingessel, et al. An Examination Of Indexes For Determining The Number Of Clusters In Binary Data Sets, Working Paper No.29, 1999. • I. H. Witten and E. Frank. DataMining: Practical machine learning tools and techniques. Morgan Kaufmann Publishers, 2005. • A.K. Jain and R.C. Dubes. Algorithms for Clustering Data. Prentice-Hall advanced reference series, Prentice-Hall. Englewood Cliffs, NJ, USA, 1988. • A.K. Jain, M.N. Murty, and P.J. Flynn. Data clustering: A review. ACM Computing Surveys, 31(3):264–323, 1999. • P.F. Velleman and D.C. Hoaglin. Applications, Basics and Computing of Exploratory Data Analysis. Boston, Mass.: Duxbury Press, 1981. • J. Quinlan. C4.5: Programs for Machine Learning. Morgan Kaufmann, Los Altos, California, 1993. • S. Haykin. Neural Networks: A Comprehensive Foundation. Prentice Hall, 2 edition, 1998. • G. John and P. Langley. Estimating continuous distributions in bayesian classifiers. Proceeding of the Eleventh Conference on Uncertainty in Artificial Intelligence, 1995.

46. Acknowledgements Dr JM Garibaldi, Dr J Bacardit Nottingham Breast Cancer Pathology RG Prof IO Ellis, Dr AR Green, Dr D Powe, Prof G Ball 46

47. Thank You! • Contact: d.soria@cs.nott.ac.uk