FpClass is an association mining algorithm that we used and validated for comprehensive, in silico prediction of physical protein interactions. FpClass is a reliable, validated method for data mining-based prediction of physical protein interactions, and provides 250,542 high confidence interactions among 10,529 human proteins, including 1,089 interactome orphans. Extensive computational and biological validation shows FpClass outperforms existing computational methods and most biological assays in sensitivity and specificity. Using three bioassays we tested 233 high and medium confidence predictions, and validated 137 interactions, including seven novel potential partners of the tumor suppressor p53.
Importantly, we validated 5 of these p53 interactions with orphans by GST pull-down assay (5 of 6 tested -- validation rate of 83%). Overall, validation rates were 40% (2/5) for co-IP, 47% (14/30) for GST pull-down, and 61% (121/198) for MaMTH (Petschnigg et al., Nat Methods, 2014). The high validation rate for MaMTH suggests that FpClass could help guide high-throughput screening, in a combined computational-experimental approach to interactome mapping. This substantially extends our interactome work, including I2D (Brown, Jurisica, Genome Biol, 2007) and (Brown, Jurisica, Bioinformatics, 2005). NAViGaTOR (Brown et al., Bioinformatics, 2009) was used for network analysis and visualization.
Supplemental table with predicted physical protein interactions