Oops, you're using an old version of your browser so some of the features on this page may not be displaying properly.

MINIMAL Requirements: Google Chrome 24+Mozilla Firefox 20+Internet Explorer 11Opera 15–18Apple Safari 7SeaMonkey 2.15-2.23

Poster display

4023 - Unsupervised latent class analysis of adult glioma variant profiles reveals biologically and clinically relevant subclasses


10 Oct 2016


Poster display


Manna Chang


Annals of Oncology (2016) 27 (6): 545-551. 10.1093/annonc/mdw393


M. Chang1, S. Ramkissoon2, H. Bokhari3, A. Fichtenholtz1, S. Ali4, J.S. Ross2, E. Neumann1

Author affiliations

  • 1 Knowledge Informatics, Foundation Medicine, 02141 - Cambridge/US
  • 2 Pathology, Foundation Medicine, Inc., 02141 - Cambridge/US
  • 3 Computational Biology, Cornell University, 14850 - Ithaca/US
  • 4 Clinical Development, Foundation Medicine, Inc., 02141 - Cambridge/US


Abstract 4023


Adult gliomas represent a diverse group of tumors characterized by differing genetic signatures and clinical presentations. Using unsupervised latent class analysis (LCA) of glioma somatic variants (independent of diagnosis), we identified biologically and clinically relevant subclasses associated with distinct clinical outcomes while highlighting the landscape of glioma driver events.


LCA was performed on somatic variants from 765 adult glioma patients (grade II-IV, TCGA dataset) using expectation-maximization and Newton-Raphson algorithms to determine maximum likelihood estimates of model parameters. Survival outcomes per class were generated using Kaplan–Meier and COX proportional hazard analysis.


LCA revealed seven distinct classes of gliomas. Classes 1 and 2 were defined by IDH1/2 mutations. Class 1 showed co-occurring CIC mutations and 1p/19q co-deletion (oligodendroglial lineage) whereas Class 2 enriched for co-occurring TP53 and ATRX mutations (astrocytic lineage). Class 3 was characterized by dysregulated cell cycle signaling largely due to alterations in MDM2/CDK4 whereas Class 4 enriched for activation of PI3K signaling mediated through alterations in NF1, PIK3CA, and PTEN. Receptor tyrosine kinase activation including EGFR and PDGFRA/KDR/KIT defined Class 5 and 7, respectively. Class 6 enriched for gliomas with gains of chr 7 but without PTEN alterations. Median survival was greatest for classes defined by IDH1/2 (2907 days Class 1, 2000 days Class 2) and poorest for Class 4 (383 days).


LCA of glioma somatic variants reveal biologically and clinically relevant classes independent of diagnosis that are associated with significant differences in patient survival. The genomic classes seem to involve different underlying molecular mechanisms that can become altered in gliomas. These findings suggest that LCA-based glioma classification may serve as a primary predictor for clinical outcomes and provide objective data to be used in clinical management and trial design.

Clinical trial identification


Legal entity responsible for the study

Foundation Medicine


Foundation Medicine


All authors have declared no conflicts of interest.

This site uses cookies. Some of these cookies are essential, while others help us improve your experience by providing insights into how the site is being used.

For more detailed information on the cookies we use, please check our Privacy Policy.

Customise settings