Fig. 4.

Unsupervised machine learning reveals changes to spontaneous motor behavior. (A) Schematic of motion sequencing approach. (B) Left, mutation plots summarizing mean usage of each syllable in wild type and mutant mice. Syllables are ordered by the relative difference between mutants and wild-type (Left: mutant down-regulated; Right: mutant enriched). Syllables with significant changes in usage, identified by Mann–Whitney U test and post hoc Benjamini–Hochberg correction, are indicated by asterisks (P < 0.05). Right, word clouds representing relative syllable changes in mutants compared to controls. Word color indicates direction of change (red: mutant upregulated; blue: mutant downregulated). Word size is proportional to the difference in usage between mutants and controls. Words are ethological descriptors assigned by reviewers. (C) Normalized classification matrices showing the performance of a classifier trained on syllable usage for mice grouped by line, sex, and genotype. An ideal classification is a value close to 1, shown in white. (D) Linear discriminant analysis plots showing similarity of mutant and wild type groups based on syllable usage signatures. Full description of statistics in SI Appendix, Table S2.