Robust and accurate deconvolution of tumor populations uncovers evolutionary mechanisms of breast cancer metastasis

Abstract

Motivation Cancer develops and progresses through a clonal evolutionary process. Understanding progression to metastasis is of particular clinical importance, but is not easily analyzed by recent methods because it generally requires studying samples gathered years apart, for which modern single-cell sequencing is rarely an option. Revealing the clonal evolution mechanisms in the metastatic transition thus still depends on unmixing tumor subpopulations from bulk genomic data. Methods We develop a novel toolkit called robust and accurate deconvolution (RAD) to deconvolve biologically meaningful tumor populations from multiple transcriptomic samples spanning the two progression states. RAD uses gene module compression to mitigate considerable noise in RNA, and a hybrid optimizer to achieve a robust and accurate solution. Finally, we apply a phylogenetic algorithm to infer how associated cell populations adapt across the metastatic transition via changes in expression programs and cell-type composition. Results We validated the superior robustness and accuracy of RAD over alternative algorithms on a real dataset, and validated the effectiveness of gene module compression on both simulated and real bulk RNA data. We further applied the methods to a breast cancer metastasis dataset, and discovered common early events that promote tumor progression and migration to different metastatic sites, such as dysregulation of ECM-receptor, focal adhesion and PI3k-Akt pathways. Availability and implementation The source code of the RAD package, models, experiments and technical details such as parameters, is available at https://github.com/CMUSchwartzLab/RAD. Supplementary information Supplementary data are available at Bioinformatics online.

Assessing the contribution of tumor mutational phenotypes to cancer progression risk

“…While some study designs might alternatively use targeted deep sequencing data, we would generally consider those data not suited to the present methods, which benefit from profiling larger fractions of the genome to estimate better aggregate mutation rates. We consider here only inference from bulk tumor data [38,39], although we note that the strategy might be applied to single-cell sequence data [40] or combinations of bulk and single-cell data [41][42][43], should such data become available for sufficiently large cohorts. The genomic data is preprocessed and passed to one or more variant callers, ideally including single nucleotide variations (SNVs) and copy number alterations (CNAs) calls as well as calls for diverse classes of structural variations (SVs) to produce a variant call format (VCF) file with detected variants and their variant allele frequencies (VAFs) per sample.…”

Semi-deconvolution of bulk and single-cell RNA-seq data with application to metastatic progression in breast cancer

“…See Avila Cobos et al (2020) for a comparison of different partial deconvolution algorithms and related data transformation methods. Examples of complete deconvolution methods include Geometric Unmixing ( Schwartz and Shackney, 2010 ), which proposed an archetype analysis method based on geometries of genomic point clouds; DSA ( Zhong et al , 2013 ), which treats deconvolution as a matrix factorization problem; LinSeed ( Zaitsev et al , 2019 ), which identifies a set of anchor genes through linear correlation and uses DSA to solve for the non-anchor genes; NND ( Tao et al , 2020a ), which poses partial deconvolution problem as a matrix factorization to be solved with gradient descent implemented through a neural network; and RAD ( Tao et al , 2020b ), which solve s the formulation of NND using a hybrid optimizer with improved accuracy and speed.…”

De novo Prediction of Cell-Drug Sensitivities Using Deep Learning-based Graph Regularized Matrix Factorization

“…2 Furthermore, responses to specific targeted therapeutics are often short-lived due to tumor heterogeneity and development of resistance. 3,4 Currently, most cancer patients are treated with “standard chemotherapies” that are not personalized, and a large proportion of patients do not respond to these therapies but suffer the full brunt of their side effects.…”