Figure 1. Overview of datasets and analyses presented in this work.

Discovery Analysis: the discovery sample was drawn from ABIDE 1. Subtypes maps and continuous subtype assignments were extracted (left, middle) from the same data and associated with ASD diagnosis (left, bottom). Replication analysis: the replication sample was drawn from ABIDE 2. Continuous subtype assignments were extracted for subjects from the replication sample, using subtypes from the discovery sample. These continuous subtype assignments were again associated with ASD diagnosis (left, bottom) to replicate the discovery findings. Stability analyses: three stability analyses were conducted, using three different datasets. Stability of discrete subtype assignments and of subtype maps (right, top) was estimated using random subsamples of the discovery dataset to regenerate the subtyping process. Stability of continuous subtype assignments in an ASD sample was estimated across scan sessions from a longitudinal subsample of ABIDE 2 (right, middle). Continuous subtype assignments were either computed for subtypes extracted from a session of the same sample (within-sample) or from the discovery sample (out-of-sample). Finally, the impact of data availability on continuous subtype assignment stability was estimated across ten scan sessions of the longitudinal HNU1 dataset (right, bottom). Subtypes were extracted from one session of the dataset, and continuous subtype assignments were computed on individual or averaged sessions (2, 3, or 4).