Fig 1.
Experimental paradigm and stimuli in (a): Forward priming and (b): Backward priming. (a): The image of galaxies (for demonstration purposes a schematic is depicted here) appeared for 1200 ms, after which the prime word was shown for 500 ms. The prime word was followed by an inter-stimulus interval (ISI) of 300 ms, where a blank was presented. After the ISI the target image (for demonstration purposes a private image is depicted here) was presented and the participant had to respond within the time duration of up to 2000 ms whether the target was of positive or negative valence. As soon a participant would press a button, the target stimulus was replaced by a blank screen, which was presented for 1000 ms and ended the trial. Backward priming trials (b) sequences were the same, with the exception of replacing the positions of target and prime with each other.
Fig 2.
Visual depiction of the neural network architecture. The input were single trial EEG traces that included all electrodes and time window of 400 ms to 700 ms after target image onset. The first layer was a convolutional layer with a kernel size matching the number of channels, filtering the data spatially. The second layer was designed to filter the data temporally. The last layer was a dense layer with a sigmoid activation function, resulting in binary classification of the trials (the class of each trial being either congruent or incongruent).
Fig 3.
(a) Median reaction times for every participant in milliseconds for the forward priming (in blue) and backward priming (in red). The x-axes depict the congruent trial reaction times and the y-axes the incongruent trial reaction times. The majority of data points in the forward priming condition are above the diagonal indicating overall shorter reaction times for congruent compared to incongruent trials. For backward priming, the data points are about equally distributed among the diagonal line, indicating no effect. (b) Box plots with difference reaction times (congruent-incongruent) for the forward and backward priming experiments. Red and blue rectangles indicate median reaction times (central horizontal line) across participants together with upper and lower quartiles. Black circles indicate median reaction times from individual participants. The forward priming box plot shows a clear negative shift, while the backward priming boxplot is close to the zero line.
Fig 4.
ERP Grand Mean (ERP) traces (congruent = dashed line; incongruent = filled-out line) for every EEG electrode. The forward priming data (a) are depicted in blue and the backward priming data (b) in red. Each subplot represents one EEG electrode. The x-axis of every subplot shows the time in milliseconds, with zero indicating target stimulus onset. The y-axis shows the amplitude in μV.
Fig 5.
Difference ERP Grand Mean (dERP) traces (congruent – incongruent ± SEM) for every EEG electrode. The forward priming data are depicted in blue and the backward priming data in red. Each subplot represents one EEG electrode. The x-axis of every subplot shows the time in milliseconds, with zero indicating target stimulus onset. The y-axis shows the amplitude in μV.
Fig 6.
Depicted are topographic maps showing median p-values for specific time windows both for the forward and backward priming experiments. The right subplot, (b), shows the backward priming p-values 207 ms to 290 ms after stimulus onset of the target image. The left subplot, (a), shows the forward priming p-values from 586 ms until 889 ms after stimulus onset of the target image. The darker the red, the less significant the electrode and the lighter the red, the more significant.
Fig 7.
Single electrode difference ERP traces.
Difference ERP traces (congruent minus incongruent) at specific electrodes at CP5 (a) and at P8 (b). The top graphs in (a) and (b) show the ERP difference traces (forward priming in blue and backward priming in red) and the bottom graphs show the uncorrected paired t-test results for every time point. The x-axes of all plots are time with zero indicating target stimulus onset (in seconds). The y-axes of the differences traces (top plots of the subplots) show the amplitude in µV while the y-axis of the t-test results (bottom plots of the subplots) shows the p-value logarithmically scaled. The black filled area of the t-test plot shows all p-values smaller than a threshold of α=0.05. Important to note is that the range of the y-axis for the t-test plots is different from (a) and (b).
Fig 8.
Averaged dERP amplitudes of individual participants.
dERP amplitudes averaged across certain time windows. Participant average amplitudes for the P8 forward priming difference trace (left) and CP5 backward priming difference trace (right) for specific time windows. Each participant is depicted in a different colour. The mean amplitude across participants is depicted by the brown square (± standard deviation). The y-axes show the single participant dERP amplitude in µ V. The P8 forward priming time window is the entire length the signal is significantly different (see Fig 7B). At the CP5 electrode, there are two time windows (TW) of the backward priming depicted. The first being the first time window where the p-value is below 0.05 at the CP5 electrode ranging from 207 to 290 ms and the second one being the second, later time window (471 to 526 ms) (see Fig 7A). For all time windows the mean amplitude was calculated and depicted here.
Fig 9.
Classification accuracies (congruent versus incongruent trials) of the artificial networks for forward (x-axis) and backward priming (y-axis) for every participant. Each icon represents the result from one participant. The brown square is the grand mean classification accuracy across all participants ± standard deviation. An accuracy value of 50% means that the network performed at chance level. An accuracy below 50% means that the network misclassified more trials than correctly classified them.