Artikel
Dose-Response Modeling for Gene Expression Data with MCP-Mod
Suche in Medline nach
Autoren
-
Julia Duda
- TU Dortmund University, Dortmund, Germany
-
Jörg Rahnenführer
- TU Dortmund University, Dortmund, Germany
| Veröffentlicht: | 26. Februar 2021 |
|---|
Gliederung
Text
Background: Traditional approaches in clinical dose-finding trials rely on pairwise comparison between doses and placebo. A methodological improvement in this field is the Multiple Comparison Procedure and Modeling (MCP-Mod) approach. While it was originally developed for Phase II trials, we broaden its usage and apply it to in-vitro gene expression data.
Methods: MCP-Mod combines the classical multiple comparisons with modeling approaches in a multistage procedure. First, for a set of pre-specified candidate models, the MCP step tests if any dose-response signal is present. Second, considering models with detected signal, the Mod step either selects the best model to fit the dose-response curve or performs model averaging.
Precisely, we apply MCP-Mod on gene expression data of human embryonic stem cells. These were exposed to different doses of the embryo-toxic, anti-epileptic compound valproic acid (VPA) at 8 dose levels with 6 (placebo) or 3 (others) replicates. Considered candidate models are the 4pLL, linear, quadratic, Emax, exponential and beta model.
Results: The analysis results give insights into the performance of the candidate models across all genes and the distribution of best-fitting dose-response shapes. Measured by the AIC, all models perform best for a considerable number of genes, but the linear model wins most frequently, roughly in one third of the cases, but mostly with no relevant performance advantage compared to the second-best model. Thus, we also investigate if one or more models can be omitted from the candidate set without resulting in unacceptable losses with respect to signal detection or target-dose estimation. We conduct a simulation study where data generation is based on the real VPA dataset and the MCP-Mod analysis is repeated with varying subsets of the full candidate model set.
Conclusion: For practical considerations, the good performance of the less used, flexible beta model is remarkable. Further, often umbrella dose-response shapes are observed.
The authors declare that an ethics committee vote is not required.
References
- 1.
- Bretz F, Pinheiro JC, Branson M. Combining multiple comparisons and modeling techniques in dose-response studies. Biometrics. 2005;61(3):738-748.
- 2.
- Krug AK, Kolde R, Gaspar JA, Rempel E, Balmer NV, Meganathan K, Jagtap S, et al. Human embryonic stem cell-derived test systems for developmental neurotoxicity: a transcriptomics approach. Archives of toxicology. 2013;87(1):123-143.
- 3.
- Pinheiro J, Bornkamp B, Glimm E, Bretz F. Model-based dose finding under model uncertainty using general parametric models. Statistics in medicine. 2014;33(10):1646-1661.
