The evaluation shows that although the model is simpler than the traditional methods, it performs well and is especially suited for situations where the initial heading and position are not known.
While it is a simple method, it performed remarkably well in experiments.
For TTkNN method, it performs worse than GkNN when k is 1, and that could be caused by noise.
We used the algorithm of Guimera and Amaral [ 35] to measure the modularity (see Methods), because it performed well in making links within modules much denser than those across modules [ 36] and has been validated in our previous network analysis [ 37].
The accuracy of FDA was comparable to state of the art approaches on both the gap gene [ 7, 8] and IRMA [ 29] data sets, which have previously been analyzed by a number of methods, and it performed well on synthetic data sets generated by the GeneNetWeaver program [ 30].
We chose the NEB normalization method since it performed best on the benchmark data H3K27me3, H3K4me3 and H3K36me3 compared to other three methods (see section Comparison of different normalization methods).
A method comprising best practices from these tests was devised, the Core Gene Similarity (CGS) method, and it performed better than simple octamer frequency analysis, codon bias, or GC contrasts in detecting seeded genes or naturally occurring transposons.
We did not calculate GO consistency scores for this method because it performed poorly for predicting genetic interactions.
Moreover, the route selection model also copes the inconsistency due to choice of different normalization methods since it performs the normalization process under AHP calculation framework.
Because Pro-Maya uses a diverse set of features, including predictions using two other methods, it also performs slightly better than other methods in the absence of additional experimental data on the query positions.
