Using their expression data and the same fold-change categories, we investigated the influence of both affinity and cooperative effects based on GraphProt predictions of Ago2 binding sites in comparison to the available CLIP-seq data.
Could do the same since expression microarray data are available, but they show complete lack of differential expression when over-expressing our proteins of interest.
Prediction margins
Part of standard GraphProt output?
logos are a mere visualization aid and do not represent the full extent of the information captured by GraphProt models
tenfold cross-validation technique
How do AUROCs look for our proteins of interest compared to the AUROCs for the iCLIP'ed proteins in Additional File 2?
The following describes a typical biological application of computational target detection. A published CLIP-seq experiment for a protein of interest is available for kidney cells, but the targets of that protein are required for liver cells. The original CLIP-seq targets may have missed many correct targets due to differential expression in the two tissues and the costs for a second CLIP-seq experiment in liver cells may not be within the budget or the experiment is otherwise not possible. We provide a solution that uses an accurate protein-binding model from the kidney CLIP-seq data, which can be used to identify potential targets in the entire transcriptome. Transcripts targeted in liver cells can be identified with improved specificity when target prediction is combined with tissue-specific transcript expression data.
use case
GraphProt: modeling binding preferences of RNA-binding proteins