Figure 1.

DTU detection demonstration

(A) Gene locus going though alternative splicing to produce isoform 1 and isoform 2.

(B) Left: isoform abundances in a sample case of DTU between individuals 1 and 2. Right: isoform abundances in a sample case without DTU but with changes in overall expression between individuals 3 and 4.

(C) Conventional DTU analysis assumption with no structured heterogeneity in either group.

(D) Heterogeneity structure in complex disease samples, where a subset of cases shares the same genetic abnormality (case-complex).

(E) Corresponding isoform fraction (IF) distributions and KDE fits for the samples represented in groups control-complex and case-complex.

(F) Three SPIT-Test iterations demonstrated with random splits of the control-complex group.

(G) Random forest regression representation when there is not a significant confounding effect in the DTU transcript (top) vs. when there is a clear confounding effect by the covariate “age” (bottom). Corresponding permutation importance scores for age and $v_j$ are shown on the right.

Samples (dots) are color coded based on their dominant isoforms for the locus in (C)–(F). Blue, isoform 1; red, isoform 2.