Repository navigation
H2 Advanced Settings
The Advanced Settings popover in the header controls how your data is processed before analysis. Click the Advanced Settings button in the top-right header to access these controls.
These settings are applied globally and affect all downstream analysis tabs.
Log-transform your intensity data before analysis. Log transformation is strongly recommended for mass spectrometry data as it stabilizes variance and makes the data more normally distributed — a requirement for many statistical methods.
Available options: log2, log10, or other bases. Log2 is the default and most commonly used in proteomics, as fold changes correspond directly to doublings (log2FC = 1 means 2× more abundant).
Missing values (features not detected in some samples) must be handled before many analyses. Available imputation methods:
- none: No imputation — missing values are left as NA
- min: Replace missing values with the minimum observed value in the dataset
- max: Replace missing values with the maximum observed value
- mixed: A combination approach
- RF: Random Forest-based imputation (most sophisticated, but slowest)
Imputation method choice can affect downstream statistics. For DIA data, missing values are typically due to low abundance, so minimum-value imputation (min) is a reasonable default.
Normalization adjusts for technical variation between samples, making them more comparable. Available methods:
- none: No normalization
- median: Median normalization — shifts each sample's distribution so medians are equal
- quantile: Quantile normalization — makes the full distribution of each sample identical
- cyclic loess: Cyclic loess normalization — iteratively fits loess curves to pairs of samples
If your samples are expected to have similar total protein content, median or quantile normalization are good starting points. If you are unsure, compare your sample distributions in the Density Plot QC tab before and after normalization.
Filter out features detected in fewer than N samples. Increasing this threshold removes low-confidence features and can improve statistical power by reducing the multiple testing burden.