For each inferred network, we computed the area under the precision-recall curve (AUPRC) relative to each of these three sets of reference networks to quantify the agreement between the predicted network and these reference networks. Similar to the BEELINE benchmarking study, when comparing inferred networks to either of the non-specific or specific ChIP-seq networks, we only considered edges outgoing from transcription factors for which ChIP-seq data was available.

In the case of our extended BVS model described in Section 2.4, we are also able to evaluate the ability of the model to accurately predict gene expression values for a network’s target genes. The linear predictor in BVS defines the relationship between each target gene and a set of regulator genes, and we evaluated the effectiveness of this predictor on unseen data by calculating the model’s posterior predictive distribution with respect to held out data. To this end, we split each cell type in a dataset into a training set comprising 80% of cells in that cell type and held out the remaining 20% of cells as a test. We trained our model on the training set and then calculated a test log likelihood value for each target gene based on the posterior predictive distribution.

Note: The content above has been extracted from a research article, so it may not display correctly.

Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.

We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.