Keywords: Machine Learning, Biostatistics, Cancer, Genetics, Public Health
Background: Early-stage prostate cancer requires normal androgen levels for growth and can be treated with androgen dependent therapy. However, malignancies inevitably advance to the castration resistant stage (CRPC) in which growth continues despite very low levels of circulating testosterone. Enzalutamide is an effective first line therapeutic treatment for CRPC that inhibits androgen receptor activity. However, most patients relapse due to the surviving tumor cells post-treatment rapidly develop resistance to enzalutamide.
Objectives: This project specifically aims to identify the genetic signature components and associated pathways with the largest functioning role for enzalutamide resistance in prostate cancer to increase targeted treatment effectiveness.
Methods: Differences in gene expression between pre-treatment and acute treatment response were analyzed using a gaussian process to model nonlinearity followed by a recently developed state-of-the-art nonlinear variable importance measure which incorporated both marginal and epistatic effects. Comparison between acute treatment response and relapse stages identified genomic drivers. Pathway enrichment of notable genes then indicated possible key biological mechanisms necessary for enzalutamide resistance.
Results: Between on-treatment and relapse, we identified many differentially expressed pathways and gene sets. These results are consistent with other published findings and include G2M checkpoints, androgen response related mechanisms, and E2F targets.
Conclusions: Comparing our findings with other studies highlights promising candidates as resistance mechanisms to enzalutamide. Lesser documented genes and pathways may be novel elements that are identified because of the incorporation of nonlinear genetic covarying relationships. These may be worth further investigation.