GPU Acceleration of a Conjugate Exponential Model for Cancer Tissue Heterogeneity

TL;DR

This work addresses cancer tissue heterogeneity by estimating subpopulation weights $K$ from gene expression data using a GPU-accelerated variational Bayes approach to a conjugate exponential model. The authors derive closed-form variational updates for $K$, $oldsymbol{eta}$, $oldsymbol{ ext{Λ}}$, and $ ho$ and implement them on GPUs with batched linear algebra, enabling scalable inference across many genes. Compared with Gibbs sampling and EM, the variational Bayes method provides faster convergence, meaningful posterior distributions, and robustness to increasing data size, demonstrated on synthetic data and real fibroblast measurements. The GPU-accelerated framework significantly reduces computation time and offers a practical pathway for analyzing tumor heterogeneity in large-scale gene-expression datasets, with potential extension to distributed systems.

Abstract

Heterogeneity in the cell population of cancer tissues poses many challenges in cancer diagnosis and treatment. Studying the heterogeneity in cell populations from gene expression measurement data in the context of cancer research is a problem of paramount importance. In addition, reducing the computation time of the algorithms that deal with high volumes of data has its obvious merits. Parallelizable models using Markov chain Monte Carlo methods are typically slow. This paper shows a novel, computationally efficient, and parallelizable model to analyze heterogeneity in cancer tissues using GPUs. Because our model is parallelizable, the input data size does not affect the computation time much, provided the hardware resources are not exhausted. Our model uses qPCR (quantitative polymerase chain reaction) gene expression measurements to study heterogeneity in cancer tissue. We compute the cell proportion breakup by accelerating variational methods on a GPU. We test this model on synthetic and real-world gene expression data collected from fibroblasts and compare the performance of our algorithm with those of MCMC and Expectation Maximization. Our new model is computationally less complex and faster than existing Bayesian models for cancer tissue heterogeneity.

Figures (12)

• Figure 1: Boolean network model of the MAPK transduction network with target locations of drugs as described in 18.
• Figure 2: The Bayesian network 25 representing dependencies of the model in h. Here $r_i$ is data, $\alpha_i$ represents subpopulation breakup, $K$ and $c$ are learnt
• Figure 3: A Bayesian network representing the conditional dependencies of our parallelizable model (subsection 3.3)
• Figure 4: Block diagram explaining the computation of the unknown parameters for our proposed model by using variational Bayes
• Figure 5: Flowchart showing the computation of the unknown parameters by using (a) MCMC, (b) variational Bayes and (c) expectation maximization method