Defining individualized theta frequency for memory modulation: A machine learning approach across brain states and regions

Recent transcranial alternating current stimulation (tACS) studies suggest that theta-frequency stimulation can modulate memory performance, with evidence highlighting individual variability in optimal stimulation frequency. However, it remains unclear which brain state ("when") and cortical region ("where") are most predictive of memory-related theta frequencies. This study aimed to identify the most relevant individualized theta frequency (ITF) parameters for episodic memory modulation using a machine learning approach. EEG data were collected from 46 healthy young-adults during rest and while performing visual (VM) and auditory (AM) memory tasks, followed by free-recall assessments. ITFs were extracted as peak theta frequencies from power spectra across 18 electrode sites and a global average ("where"), across three states: resting, task-encoding, and task-delay ("when"). Participants were clustered into high- and low-performing groups based on ITFs using K-means clustering, and candidate ITFs were further examined via correlation and Bayesian regression analyses to assess their predictive power. All ITF candidates showed some clustering success, but global task-state ITFs best distinguished between performance groups, independent of task modality. Notably, resting-state left posterior parietal (LPP) ITF was negatively correlated with both VM and AM performance, suggesting a domain-general role in baseline memory capacity. Additionally, task-specific contributions were observed: encoding-related left temporoparietal and delay-related left central ITFs were significantly associated with AM performance, potentially reflecting auditory-specific processes. These findings highlight the importance of "when" and "where" specificity in defining individualized stimulation protocols. Resting-state LPP ITF, in particular, may serve as a promising biomarker for tailoring tACS at sub-ITF frequencies to enhance memory performance.