Functional transcriptional signatures for tumor-type-agnostic phenotype prediction

Abstract

Gene expression predicts tumor characteristics such as resistance to anticancer therapy. However, generalizing these predictors to multiple cancer types and data sets to motivate new therapeutic strategies has proven difficult. Here, we present a nonnegative matrix factorization (NMF) approach that decomposes gene expression into a universal set of archetype fingerprints. By restricting our analysis to five well-defined biological pathways, we show that trade-offs between normal tissues constrain oncogenic heterogeneity. Thus, the resulting six archetypes unify gene expression variation across 54 tissue types, 1504 cancer cell lines, and 1770 patient samples. The archetype mixtures correlate with cancer cell line sensitivity to several common anticancer therapies, even among cancers of the same type. They also explain subtype-specific breast cancer characteristics and define poor prognostic subgroups in breast, colorectal, and pancreatic cancers. Overall, the approach offers an evolvable resource for understanding commonalities across cancers, which could eventually lead to more robust therapeutic strategies.

Mapping the Single-cell Differentiation Landscape of Osteosarcoma

“…The archetypes accurately quantified the presence of various differentiation states and lineages in human OS samples. They also showed the potential to quantify the impact of suspected driver alterations and transcription factors on tumor phenotype 15 . Given the link between these archetypes and patient survival, this approach could aid in the development of patient-specific therapeutic strategies for OS.…”

“…We applied a Normalized Non-negative Matrix Factorization (N-NMF) algorithm to define transcriptional programs or “archetypes” which capture the variability of gene expression patterns across the dataset in a low-dimensional subspace 15 . This approach is a semi-supervised machine learning approach, where we first trained N-NMF archetype coefficients on the mesenchymal differentiation dataset and later used these trained archetypes to score the osteosarcoma PDX and tumor datasets.…”

“…We hypothesized that the MTL would catalog the various differentiation states possible in OS. To quantify differentiation states in the MTL, we applied a Normalized Nonnegative Matrix Factorization (N-NMF) based archetype analysis to identify recurring gene expression profiles that accurately captured lineage-specific temporal dynamics 15 . These profiles were then used to estimate the relative compositions of each cell lineage and differentiation status within OS tumor data, including patient-derived xenograft (PDX) and tumor biopsy samples 16,17 .…”