Frequently Asked Questions for DoseRider
What are the different results in the results table?
The results table provides a comprehensive set of metrics for each pathway model:
- P-value (Linear / Non-linear Fixed / Non-linear Mixed): These p-values indicate the statistical significance of the pathway response under different models - linear, non-linear fixed effects, and non-linear mixed effects models. We test nested models to see if each subsequent model explains the variance better than the previous one. The comparisons are Null vs. Linear, Non-linear Fixed vs. Linear, and Non-linear Mixed vs. Non-linear Fixed. We aim to keep the most simple model that best explains the data.
- Adjusted P-value: This is the p-value adjusted for multiple comparisons to control the false discovery rate.
- AIC (Akaike Information Criterion): AIC is used for model selection in regression analysis, especially when the sample size is large. Lower values of AIC indicate a better fit of the model to the data.
- AICc (small-sample corrected Akaike Information Criterion): AICc is similar to AIC but includes a correction for small sample sizes. When the sample size is large, AICc converges to AIC, making them very similar. Lower values of AICc indicate a better fit of the model to the data.
- BIC (Bayesian Information Criterion): BIC is another criterion for model selection with a different penalty for the number of parameters compared to AIC. Lower values of BIC indicate a better fit of the model to the data.
- df: Degrees of freedom for each model, indicating the number of independent parameters in the model.
- Optimal Clusters: The number of clusters that best explain the data. The optimal number of clusters is determined using the silhouette method based on dose-response gene expression data, where the silhouette width metric is used to evaluate the quality of the clustering.
What types of data can DoseRider analyze?
DoseRider is optimized for analyzing dose-response data from chemical, drug, and environmental exposure experiments involving varying doses. It expects "expression" or "count" data where columns are samples and rows are features such as genes, proteins, lipids, etc. Additionally, a metadata file is required, where rows are the samples and each column is a different feature. The metadata must include a "dose" column, indicating the doses used in each sample for dose-response modeling. Supported data formats include CSV, Excel, and certain database exports.
How does DoseRider perform pathway modeling?
DoseRider uses generalized mixed models to perform pathway modeling. Each individual gene within a pathway is considered an observation of the pathway. All genes within a pathway are used to model the dose-response relationship, providing a comprehensive analysis of pathway responses.
What statistical methods and testing does DoseRider employ for analyzing data?
DoseRider supports various models to fit the pathway response, including:
- GLM (Null model): No dose-response relationship (baseline model).
- Linear mixed model: Linear dose-response relationship (dose is a fixed effect, gene is a random effect).
- Non-linear fixed model: Non-linear relationship using cubic splines for fixed effects.
- Non-linear mixed model: Adds gene-specific cubic splines.
To select the best model, DoseRider follows the recommendations from the US EPA guidance, considering both p-values and AIC for model fitting.
How does DoseRider compute the Benchmark Dose (BMD)?
DoseRider computes the BMD using the BMDz method. This function calculates the BMD based on a smoothed trend from model predictions. The BMD is identified as the dose level where the predicted response exceeds a threshold, defined as a specified number of standard deviations (‘z’) above the control response. It is designed to work with smoothed trend data, typically derived from 'lmer' or 'glmer' model predictions.
Can DoseRider analyze multi-omics data?
Yes, DoseRider can work with multi-omics data but only fits one type of data at a time. It selects the best data distribution, either Gaussian or Negative Binomial, and estimates the necessary parameters when required.
What species are available for study with DoseRider?
The main limitation for working with different species is the geneset annotation. Currently, DoseRider includes genesets from Homo sapiens, sourced from ConsensusPathDB, MsigDB, and MIO (see references). Efforts are ongoing to include more GMT files from different species and to allow users to upload their own datasets.
Does DoseRider provide visualization of dose-response curves?
Yes, DoseRider offers comprehensive visualization tools that enable users to generate and customize dose-response curves. Visualization options include logarithmic and linear displays, along with various curve-fitting parameters.
Is DoseRider free to use?
Yes, DoseRider is a free online tool available to anyone with an internet connection. We request that users cite our software in their publications.