Caveats

As with any tool in biology, the information in TransfactomeDB should not be interpreted in isolation. However, when the information in the database is combined with knowledge gleaned from the original microarray publications and other literature sources, the user can derive specific and meaningful conclusions. Moreover, the database can be used to generate experimentally testable hypotheses about transcriptional regulation (DBPs) and post-transcriptional regulation (RBPs). Of course, it inherits any limitations of the original data from which it is derived. First, not all microarray experiments in GEO contain meaningful data or necessarily measure only the phenomenon that the experiment was designed to measure. Second, a PSAM derived from genome-wide occupancy data is not necessarily the PSAM for the named, assayed factor or even a real PSAM. We report the best fit PSAM for each ChIP-chip or similar experiment. Therefore, if the original occupancy data is due to the specificity of another factor or is completely noise, then the reported PSAM will not accurately reflect the binding specificities for the factor of interest. The t-value for the goodness of fit for the PSAM to the occupancy data can provide a measure of whether the PSAM is derived from pure noise. To mitigate uncertainty in DBP identity, the database allows the user to optionally list only those PSAMs that have a similar predicted occupancies as PSSMs from MacIsaac et al. [5] or PWMs from TRANSFAC [6] for the same DBP or related DBPs. However, there is no computational solution to perfectly prevent reporting PSAMs for factors that are physically interacting with to the immunoprecipitated factor in ChIP-chip experiments. Only data generated from in vitro genome-wide occupancy experiments [18,19,20,21] can assure that the microarray signal and thus the PSAM is due to the assayed factor. Finally, a high scoring PSAM match for the "Dissect Sequence" tab can only imply potential regulation. This feature can be powerful if combined with previous knowledge and experimental validation. However, the user should expect frequent false positives.