BIOPTRAIN Meeting

1
BIOPTRAIN Meeting

• Daniele Soria

2
Who I am

• Daniele Soria (Room B79)
• Bioinformatic analysis of breast cancer data
• First Supervisor Dr. Jon Garibaldi

3
Main goals

• Collaboration between U.o.N. and I.N.T.
• Combination of various data sets available at the
  centers into a single European resource for
  breast cancer data
• Investigation of novel computational analysis
  methods applicable across bioinformatics data and
  routine clinical information

4
How to reach the goals

• SQL data management, in order to work better with
  databases
• Meetings with pathologists and researchers of the
  School of Molecular Medical Sciences at
  Nottingham City Hospital

5
Current work (1)

• Databases management
• Discrepancies in different clinical databases
  from QMC also involving Gene Microarray data and
  Mass Spectrometry data
• Unique dataset with all information available
  (still in progress).

6
Current work (2)

• Analysis of Breast Cancer populations involving
  various centers and different steps
• 1. Cluster definition Fuzzy C-means Clustering
  (UoN), Hierarchical Euclidian (NTU), K-means
  and Adaptive Resonance Theory (Liverpool
  University), other methods (Milan)
• 2. First statistical analysis data description,
  cluster comparison and evaluation in terms of
  clinical characteristics (Kaplan-Meier of
  survival, disease free interval and time to
  recurrence)

7
Databases management

• 15 different databases provided by Nottingham
  City Hospital and Nottingham Trent University
• 2500 cases with almost 300 variables (clinical
  data and biomarker determinations)
• Find discrepancies between data and check all
  possible duplications
• Put all together

8
Analysis of BC populations

• Fuzzy C-means clustering
• Minimization of the objective function
• X x1,x2,...,xn collection of n data points,
  V v1,v2,,vc set of corresponding c
  cluster centres in the dataset X,
• µij membership degree of data xi to the
  cluster centre vj
• m (fuzziness index) used to control the
  fuzziness of membership of each datum (often m2)

9
Analysis of BC populations

• Xie-Beni validity index
• Evaluation of the quality of clustering
• May be used to find the optimal number of
  clusters
• Measure of clusters compactness and separation
• If S represents the overall validity index, is
  the compactness and s is the separation of the
  fuzzy c-partition of the dataset, then Xie-Beni
  validity index can be expressed as

10
First results (1)

• Split of data in two clusters
• Calculation of Xie-Beni validity index for all
  clustering methods used
• Visualisation of Kaplan-Meier curves

11
First results (2)
12
First results (3)
13
Next steps

• Chi-squared tests on survival between clusters
  cut-off 5 years without recurrence
• Mann-Whitney Test on continuous variables divided
  by cluster groups
• Non-parametric statistical models (Cox)

14
Acknowledgments

• Dr. J. Garibaldi
• I.N.T. Dr. E. Biganzoli, Dr. P. Boracchi
• Dr. A. Green, Dr. G. Balls
• Marie Curie Action MEST-CT-2004-7597 6FP of
  European Community
• Other PhD students (especially Xiao Y Wang)

15
The End Thank you!

• World Champions
• ITALIA