Speaker
Description
Patients with the same cancer diagnosis often respond very differently to therapy, motivating efforts to identify molecular subgroups explaining this heterogeneity. High-dimensional molecular datasets are often analysed using correlation-based similarity among features\cite{1,2}; however, correlation captures global expression patterns and often fails to reveal how signals split patients into distinct molecular states. We present Partition-Induced Signal Analysis (PISA), a framework that groups bimodal molecular signals by the patient partitions they induce, defining feature similarity via agreement of induced splits rather than expression correlation. Bimodal features, where samples cluster around two expression states, naturally define candidate patient splits\cite{3}. Such signals are identified using density criteria. Gaussian mixture models estimate partitions induced by each feature, and features producing concordant splits are aggregated into coherent signals. Applied to breast cancer data from the I-SPY 2 Trial, PISA uncovers five latent patient profiles from gene expression data, several supported by phosphorylation patterns in RPPA data. Stratification by these profiles improves prediction of therapeutic response relative to phosphorylation-based biomarker rules\cite{4} and suggests patient groupings that may increase response rates beyond conventional subtype stratification\cite{5}, revealing hidden biological structure enabling more precise patient stratification.
Bibliography
@article{2,
title={A general framework for weighted gene co-expression network analysis},
author={Zhang, Bin and Horvath, Steve and others},
journal={Statistical applications in genetics and molecular biology},
volume={4},
number={1},
pages={1128},
year={2005}
}
@article{1,
title={Cluster analysis and display of genome-wide expression patterns},
author={Eisen, Michael B and Spellman, Paul T and Brown, Patrick O and Botstein, David},
journal={Proceedings of the National Academy of Sciences},
volume={95},
number={25},
pages={14863--14868},
year={1998},
publisher={The National Academy of Sciences}
}
@article{3,
title={The bimodality index: a criterion for discovering and ranking bimodal signatures from cancer gene expression profiling data},
author={Wang, Jing and Wen, Sijin and Symmans, W Fraser and Pusztai, Lajos and Coombes, Kevin R},
journal={Cancer informatics},
volume={7},
pages={CIN--S2846},
year={2009},
publisher={SAGE Publications Sage UK: London, England}
}
@article{4,
title={Protein signaling and drug target activation signatures to guide therapy prioritization: Therapeutic resistance and sensitivity in the I-SPY 2 Trial},
author={Gallagher, Rosa I and Wulfkuhle, Julia and Wolf, Denise M and Brown-Swigart, Lamorna and Yau, Christina and O’Grady, Nicholas and Basu, Amrita and Lu, Ruixiao and Campbell, Michael J and Magbanua, Mark J and others},
journal={Cell Reports Medicine},
volume={4},
number={12},
year={2023},
publisher={Elsevier}
}
@article{5,
title={Redefining breast cancer subtypes to guide treatment prioritization and maximize response: Predictive biomarkers across 10 cancer therapies},
author={Wolf, Denise M and Yau, Christina and Wulfkuhle, Julia and Brown-Swigart, Lamorna and Gallagher, Rosa I and Lee, Pei Rong Evelyn and Zhu, Zelos and Magbanua, Mark J and Sayaman, Rosalyn and O’Grady, Nicholas and others},
journal={Cancer cell},
volume={40},
number={6},
pages={609--623},
year={2022},
publisher={Elsevier}
}
