Human brain integrates both unconditional and conditional timing statistics to guide expectation and behavior

Our brain uses prior experience to anticipate the timing of upcoming events. This dynamic process can be modeled using a hazard function derived from the probability distribution of event timings. However, the contexts of an event can lead to various probability distributions for the same event, and it remains unclear how the brain integrates these distributions into a coherent temporal prediction. In this study, we create a foreperiod sequence paradigm consisting of a sequence of paired trials, where in each trial, participants respond to a target signal after a specified time interval (i.e., foreperiod) following a warning cue. The prediction of the target onset in the second trial can be based on two probability distributions: the unconditional probability of the second foreperiod and its conditional probability given the foreperiod in the first trial. These probability distributions are then transformed into hazard functions to represent the unconditional and conditional temporal predictions. The behavioral model incorporating both predictions and their mutual modulation provides the best fit for reaction times to the target signal, indicating that both temporal statistics are integrated to make predictions. We further show that electroencephalographic source signals are also best reconstructed when integrating both predictions. Specifically, the unconditional and conditional predictions are encoded separately in the posterior and anterior brain regions, and integration of these two types of predictive processing requires a third region, particularly the right posterior cingulate area. Our study reveals brain networks that integrate multilevel temporal information, offering insight into the hierarchical predictive coding of time.