The human phosphoproteome was constructed by combining Ochoa et al.'s high-throughput data with kinase-substrate data from multiple sources. KMeans clustering was performed using 50, 150, 300, and 500 clusters across three random seeds. Clusters were merged based on a Jensen-Shannon divergence (JSD) threshold of 0.05 to reduce redundancy. Clusters with fewer than 40 unique sub-site IDs or a maximum surrounding PSSM value below 0.4 were filtered out.