Fig. 2: Reconfiguration of the low-dimensional trajectories of brain activity as a function of working memory load, task performances, and local neural perturbations.
a A shared (common) embedding space allows an across-session comparison of the state-space embedding of trajectories of brain activity supporting “correct” and “incorrect” trials. A general expansion of the trajectories of activity (i.e., utilizing data from all experimental sessions) is required to support “correct” trials in the 2-back condition (2.6 times greater than 1-back trials and 3.4 times greater than ‘incorrect’ trials in the same conditions). b Analysis of ‘correct’ and ‘incorrect’ trials for the iPS session showed that a significant expansion of low-dimensional trajectories of brain activity following a targeted neural perturbation (TMS on iPS; baseline and S1 shown in Supplementary Fig. 4a) supported successful 2-back performance (5.4 times greater than 1-back trials and 8.1 times greater than “incorrect” trials). For panels (a and b), asterisks indicate significant (*pFDR < 0.001) and highly significant values (**pFDR < 1 × 10−6). c The significant expansion of trajectories following TMS on iPS involved the three PHATE dimensions (‘correct’ trials in the 2-back condition, visualized in log space, same embedding space across experimental sessions). d At the participant-level (n = 17, examined over three experimental sessions), comparison of the lengths of trajectories supporting distinct working memory loads (2-back minus 1-back, subtracting ‘incorrect’ trials; median represented by red line; within-session S.D. shown with vertical black lines) showed that “correct” trials following TMS on iPS were linked to a greater expansion of the low-dimensional trajectories as a function of load compared to baseline and S1 trajectories. Comparisons of trajectory lengths across sessions were tested with two-sided paired t-tests, with significance values corrected for multiple comparisons. Source data are provided as a Source data file.
