What each plot is showing
The math behind these lives in the Methodology page. This page is about reading them.
The axis scatter (2D / 3D)
Each point is one sample, placed by its position in the model's learned latent space, a compressed representation of that sample's expression profile with dataset and tissue-site effects suppressed. Points that sit close together are biologically similar in a way the model considers meaningful, independent of which cohort or exact tissue site they came from. Colour reflects the selected axis, so switching axes re-colours the same underlying positions rather than moving the points.
The radar profile
A shape-based summary of one selected sample's calibrated pathway scores across every axis at once. It's built for fast visual comparison. Two samples with similar radar shapes have a similar overall pathway activity signature, even if you haven't looked at the individual numbers.
The pathway heatmap
The same per-axis scores as the radar view, but as exact standardized values rather than a shape. Use this when you need the precise number for a specific pathway rather than a general impression.
Reading the pathway scores themselves
Pathway scores are latent decoder-derived pathway coordinates, standardized against the internal IBDex reference distribution. They are not ssGSEA scores, and shouldn't be compared directly to ssGSEA output from another tool. A value near 0 means "typical for the reference cohort"; larger positive or negative values mean further from that typical range in either direction.