To assemble the broadest possible data sets to test the approach and make predictions, we take all known interacting proteins assembled in a published data set that contains ∼3500 high-confidence protein interactions in E. coli (Rajagopala et al., 2014).
To report posterior probabilities for phylogenetic footprinting predictions we take the approach of [ 4, 5, 20], collapsing our predictions onto one axis and reporting posterior probabilities for a single species.
If there was no significant hit for any of overlapping predictions, we took the longest one.
While perfoming these predictions, we took tRNA genes from the genome sequence of S. japonicum UT26S, whose total genome sequence is known (Nagata et al. 2010).
To evaluate the accuracy of enhancer predictions, we took into account the collection of 1 830 experimentally validated CRMs published in REDfly v3.2 [ 39].
To support EST-driven predictions, we took advantage of a previously described experimental model, which is based on the interaction between A. thaliana and its compatible pathogen, yellow strain of Cucumber mosaic virus (CMV (Y)) [ 5].
To search for possible new alternatively spliced exons in the remaining 71,300 Exoniphy predictions, we took all such predictions that satisfy the following criteria: (i) Found within an intron of a known RefSeq gene (ii) Do not overlap with any EST/RNA, (iii) Do not overlap with any processed pseudogene, and (iv) Are flanked by canonical splice sites both in the human and the mouse genomes.
In the case of prediction, we take x k ≈ x k k − 2, so H ( x k, x 0 ) ≈ H ( x k k − 2, x 0 ).
In order to compute equation (4), we approximate the second-degree term H ( x k, x 0 ) by using the most current available state estimation for x k ; that is, In the case of prediction, we take x k ≈ x k k − 2, so H ( x k, x 0 ) ≈ H ( x k k − 2, x 0 ).
Moreover, in order to automatically build the classification model for prediction, we take two snapshots of the concept networks corresponding to two consecutive time durations, such that a training data set can be formed based on a group of labeled concept pairs that are automatically extracted from the snapshots.
To be able to calculate the performance of the voting prediction, we took into consideration only the users who expressed vote intention for a specific party.
