• Predictions made by integrating information from protein-protein interaction assays, including both reliable ones and high-throughput ones.
• Excludes motifs whose length covers more than half of the protein.
• Excludes proteins that have zero probability to contain any driver mutation.
• Preditions high on the list are more likely to be the actual binding sites, while predictions near the bottom are not likely to be the binding sites.

See original paper for more details.