Details of the validation of cluster-based filtering

Rank these epitopes based on their cognate TCR counts in descending order. The ranking is expressed in percentiles, where lower percentile values represent epitopes with higher cognate TCR counts, implying a higher frequency or prevalence of these epitopes within the dataset. Conversely, higher percentile values indicate epitopes with lower cognate TCR counts, suggesting these epitopes are less frequent. The distribution of percentile values for these epitopes is illustrated in the 'Before' category in Figure 4B, with the values representing the median.

After filtering the dataset, we plotted the percentile values of all remaining epitopes using violin plots, keeping the percentiles consistent with those obtained in the previous step. For example, before filtering, the dataset had six epitopes[A, B, C, D, E, F] with percentiles [0, 0.2, 0.4, 0.6, 0.8, 1.0]. After filtering, the dataset retained only three epitopes [A, C, E], and the percentiles for these epitopes remained [0, 0.4, 0.8]. These are shown in the 'After' category in Figure 4B.

Figure 4B helps illustrate changes in the distribution, including any shifts in the median and alterations in the distribution's spread (as indicated by the waist of the violin plots).