Harnessing Diversity towards the Reconstructing of Large Scale Gene Regulatory Networks
Figure 1
Example of a prediction by TopkNet formed from five individual network predictions.
(a) Target Network. Circles and links are genes and regulatory links among genes, respectively. (b) The five lists are ranked according to the confidence levels of links, the most reliable prediction is at the top of the list and has the highest rank, i.e., Algorithm1 assigns the highest confidence level and the rank value of 1 to a link between nodes 1 and 2. The true link of the target network is highlighted in yellow. We regard links with rank of 1–7 as regulatory links inferred by the algorithms because the target network composed of 7 links. Red lines and blue dashed lines represent true positive and false negative links, respectively. (c) Five rank values for each link and the mean value among the five values. Green, red, orange, blue, and purple represent rank values from Algorithm1, Algorithm2, Algorithm3, Algorithm4, and Algorithm5, respectively. (d) Rank value of a link by TopkNet and that by Community Prediction. Top1Net and Top2Net regards 1st and 2nd highest value among five rank values for a link as the rank value of the link, respectively. Community Prediction calculates the mean value among five rank values for a link and regards the mean as the rank of the link. For example, rank of the links between genes 1 and 2 for Community Prediction is 7.4. This example illustrates how Top1Net can be more accurate than the other algorithms.