1 / 13

Sohrab Shah Department of Computer Science University of British Columbia

Detection of structural abnormalities in tumour genomes using model based approaches: application to 107 patients with follicular lymphoma. Sohrab Shah Department of Computer Science University of British Columbia. BC Cancer Research Centre : Doug Horsman K-John Cheung Jr.

paiva
Télécharger la présentation

Sohrab Shah Department of Computer Science University of British Columbia

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Detection of structural abnormalities in tumour genomes using model based approaches: application to 107 patients with follicular lymphoma Sohrab Shah Department of Computer Science University of British Columbia BC Cancer Research Centre: Doug Horsman K-John Cheung Jr. UBC computer science: Kevin Murphy Raymond Ng

  2. Structural abnormalities in cancer Detection of structural abnormalities in tumour genomes using model based approaches

  3. Amplification Nature437, 1084-1086 Copy number alterations (CNA) can lead to disease Bayani et al, Cancer Research 2002 • CNAs are a hallmark of tumor genomes • CNAs can lead to adverse expression changes of affected genes • Recurrent CNAs in patients with common phenotype potentially represent molecular markers of disease • Task: find recurrent CNAs for diagnostics, gene-disease association, disease susceptibility Detection of structural abnormalities in tumour genomes using model based approaches

  4. Neutral Gain CNA labels? Loss Goal: Classify each probe as loss, neutral, gain Solution: fit a hidden Markov model (HMM) to the data Detect CNAs using array comparative genomic hybridization (aCGH) 27K probes Per patient aCGH data Detection of structural abnormalities in tumour genomes using model based approaches

  5. Why HMMs for aCGH? Ground truth labeled data Advantages of an HMM: • Student-t mixture emission model • HMM transition matrix • Continuous data -> discrete biology • measurement noise • spatial correlation • Classification (L,N,G) Detection of structural abnormalities in tumour genomes using model based approaches

  6. Our HMM leads to improved accuracy • Contribution: novel HMM adaptation for aCGH • Extension of Fridlyand et al (2004) • 15% improvement over state of the art • 95% classification accuracy for 49 manually annotated samples • Shah et al, Bioinformatics (2006) Detection of structural abnormalities in tumour genomes using model based approaches

  7. Large-scale study of follicular lymphoma (FL) • 107 patients, aCGH data: 27K probes per patient • Manual annotation of all patients • Clinical data available • Survival • Time to transformation to more aggressive stage • GOAL: provide a pattern of recurrent CNAs (called a profile) that characterize this disease • Pick specific probes for validation • Determine affected genes Multiple aCGH samples CNA profile Detection of structural abnormalities in tumour genomes using model based approaches

  8. Analysing 107 aCGH profiles of follicular lymphoma Neutral Gain Patients Loss Probes Detection of structural abnormalities in tumour genomes using model based approaches

  9. 1p36: region of interest • Experimental validation rate of 79% using FISH Alteration frequency (AF) vs manual – Chr 1 Loss Manual AF Loss AF Gain Where are the signals strongest? Gain Detection of structural abnormalities in tumour genomes using model based approaches

  10. A novel Hierarchical HMM (HHMM) for inferring recurrent CNAs Borrow statistical strength across patients using raw data Focus on consensus Explicit modeling of ambiguity distinguishes ‘random’ effect from shared signals Produce sparse output where signals are strongest Shah et al Bioinformatics 2007 Detection of structural abnormalities in tumour genomes using model based approaches

  11. HHMM yields sparse output where shared signals are strongest Loss Manual AF Loss AF Gain HHMM HHMM-U Gain Detection of structural abnormalities in tumour genomes using model based approaches

  12. Future work • Validation of HHMM predictions on an independent cohort • Extend HHMM for clustering patients • Stratification of the population based on aCGH may point to distinct molecular subtypes of FL • Correlation of sub-groups to clinical variables may lead to prognostic profiles • Detect subgroup-specific markers that are distinct from ‘background’ • Clinically predictive markers? Detection of structural abnormalities in tumour genomes using model based approaches

  13. Acknowledgements Advisors: Kevin Murphy and Raymond Ng Collaborators: K-John Cheung Jr., Douglas Horsman Michael Smith Foundation for Health Research: Senior graduate scholarship Genome Canada/Genome BC: Research grant for array CGH http://www.cs.ubc.ca/~sshah Detection of structural abnormalities in tumour genomes using model based approaches

More Related