Sometimes there is a beautiful timing in how things arrive in the literature. For instance with my collaborative work for CDD I have been writing manuscripts on how we can use machine learning models to predict likely actives in whole cell phenotypic screens for M. tuberculosis. I have tested these models, retrospectively, prospectively, within the same lab, across labs, you name it..Then all of a sudden yet another lab, in this case GSK, publishes a high throughput screening study after screening 2 million compounds. Yes you read correctly, even though there are computational models from us and elsewhere (for years) they still screened a whole library..nevermind. The dataset of 11 compounds in the paper has been a useful additional test set – results to be revealed in due course…Back in 2010 we also published predictions for a subset of the GSK library when they released malaria actives, so it would be interesting to see how they worked out but this will take a little digging.
Second the dataset of 11 molecules they share in the paper is useful as a test set for the TB Mobile app to predict potential targets. The app is available for Apple and Android devices for FREE. So a few hours have been spent doing a little open science. First I sketched the 11 molecules in MMDS from Alex Clark (this took about 30-40 min), then I exported each molecule into TB Mobile and performed the similarity search (this took about 20 minutes). I simply took a screengrab from the iPad and then annotated the target of the top couple of molecules from the TB mobile app. Putting this all together on a powerpoint poster took much longer. So what is missing? Obviously it would be valuable to validate the predictions for each compound in vitro etc. and that will take far more than a few hours of work on my iPad. For the wet lab validation it is unclear how readily available these 11 compounds will be or whether they are commercially available. If I was a compound vendor I would take note too.
While the similarity approach is far from definitive, it may be instructive. This example also represents one of the first attempts at personally doing a workflow in a mobile app. While I used MMDS (which is not free – but I got a freebie from Alex) to handle a spreadsheet and draw the molecules, the sketching of molecules in TB Mobile is possible one molecule at a time. MMDS was very useful because I had 11 molecules and I wanted to produce a table with the predictions alongside compound ID information that was publication quality for the poster (exported to excel). All in all I found doing the parts on the iPad easy, drawing structures, using apps. The trick comes in then putting the outputs together as a poster and then blogging here -I reverted back to my PC.
I hope this inspires others to try out the mobile apps, and if BTW you are thinking about doing any HTS for M. tuberculosis, please get in touch as I am more than happy to share the TB Bayesian Models in the interests of open science and collaboration. If anyone wants to test the compounds..good luck and please let me know how you get on.
NOTE added after blog posted – See comment below Chris Southan also blogged on this in the past week relating to the GSK compounds . great minds think a like, I am just a few days late!