Figure 1

Architecture of the predictor and likelihoods of the modules. The predictor consists of two different parts (A and B) which are trained consecutively. The Group A modules (shown in panel A) are trained in parallel. The likelihood ratios (LR) for most of their states are shown in panel A (their complete likelihood ratios are available in Additional File 4). The product of the likelihood ratios of all Group A modules considered in a given prediction is referred to as the preliminary score (PS) and can be calculated for all human protein pairs. If the Transitive module is not considered, the final likelihood ratios assigned to all protein pairs is the preliminary score (PS). If the Transitive module is considered, the local topology of the network determined by the assignment of preliminary scores to all protein pairs considered in the training set is used to calculate the likelihood ratios for the transitive module (shown in panel B) for every protein pair in the training set. The final likelihood ratio is then the product of the preliminary score calculated in panel A and the likelihood ratio output by the transitive module in panel B. For the Orthology module: YL, YM, YH: yeast low, medium and high scoring bins; FL, FM, FH: fly low, medium and high scoring bins; WL, WM, WH: worm low, medium and high scoring bins; HM and HL: medium and low scoring bins for human protein pairs that have human paralogs; > 1 organism: bin for human protein pairs that have interologs in more than one organism. For the Combined module, –— refers to the lowest scoring bin (for the domain (Dom), post-translational modification (PTM) and subcellular localization (Loc) features), – refers to the second lowest scoring bin and +, ++, +++ refer respectively to the third highest, second highest and highest scoring bins.