SwitchClass distinguishes baseline- and perturbation-aligned features across intermediate molecular states.

Biological systems often exhibit intermediate molecular states under perturbation, showing changes that align with baseline condition or perturbed state. Capturing these complex patterns is critical for understanding molecular resilience and maladaptive persistence. We introduce SwitchClass, a label-switch classification framework that distinguishes features whose intermediate-state profiles align with baseline or perturbed condition. By training a dual classifier with inverted labels, SwitchClass computes a directional importance score (δ), which quantifies each feature's alignment across biological states. Applied to colorectal cancer proteomics, SwitchClass reveals proteins that normalize after therapy and those remaining dysregulated, uncovering partial molecular recovery. In dietary perturbation and reversal phosphoproteomics, it uncovers the phosphorylation sites linked to incomplete restoration of insulin signaling. In single-cell transcriptomes from COVID-19 patients with varying severities, it identifies cell-type-specific transcripts marking resolution or persistence of inflammatory activity. Together, these analyses demonstrate SwitchClass as an interpretable framework for mapping directional molecular changes in systems with intermediate states.