ChEMBL

Having just moved to a new office, it is, of course, the best time to sort through all the stuff you have just moved to decide what to keep (note to self - next time throw out rubbish before moving). I found a whole pile of old ideas, sketches, hard-copy presentations that no longer exist in digital form, and of course most of it/all of it is rubbish.

One thing I reflected on though, was a scribbled note on an "Amazon for SAR data" (unsurprisingly, the page had a list of potential names, the strongest (relatively) was SARmazon). The basic idea was to use search preferences to suggest jumps to other data that may be of interest, essentially just a series of links between papers (orderable by date, number of compounds, etc.) The links would be established by compound similarity and target/assay similarity. More specifically you would use the shared occurrence of compounds, or targets, to build the association list and weights. It would be a little like the systems that on-line shops use to suggest purchases, based on what others (estimated as having similar tastes to you) have purchased. Its just here you use compounds as the objects in the store, and the papers/patents as the purchasers. Anyway, having just read this explanation, it is not really clear, and an example may make it so.

Anyway, I have just revisited this idea, using our latest StARlite data, and it looks quite interesting (for some cases, and uninformative for others, but hey, sometimes Amazon suggests the Pet Shop Boys, and sometimes Franz Ferdinand - PSB everytime!). Watch this space....