ChEMBL Resources

Resources:
ChEMBL
|
SureChEMBL
|
ChEMBL-NTD
|
ChEMBL-Malaria
|
The SARfaris: GPCR, Kinase, ADME
|
UniChem
|
DrugEBIlity
|
ECBD

Saturday, 15 September 2012

Query Privacy in ChEMBL


We have been asked several times for all the user-generated queries of ChEMBL - i.e. the structures sketched in to the interface that are then searched against the database. We will not (and in fact, physically can't) share these. Sorry. It is against both our institutional privacy policy, and standard Terms of Use, and also we've engineered the app to avoid us 'storing' any of this information where at all possible (e.g. in avoiding /tmp type fluff, minimizing residency time in caches, etc.).

There are clearly some advantages in pooling or analysing website search data - it highlights interesting trends, something becoming more interesting to a user community can spot emerging events, etc. It can alert to flu outbreaks (there was a Science paper from google on this, don't have the reference handy though - you may be able to find it with google though.....). There is a huge interest in many sites that I use in tracking and analysing query terms and usage patterns, and in some contexts this is just the thing to do - like when ebay teases me (and surely of all the tortured obsessive souls on the planet, it is just me and me alone) with a rare phosphor or perforation machin variant I don't have.

The types of query that people perform can clearly also be used to develop ways of improving a website, or specifically the performance of search queries - and for algorithm development this information can be like gold-dust. There are now many chemical fingerprint systems available, and adapting the features/structures of these to typical user queries is really valuable in their development.

There are essentially two distinct aspects to user's expectations/rights of privacy when using a website like ChEMBL.

  • There is a personal privacy issue - 'why is John Overington interested in compounds for the treatment of obesity?'. This is an primarily an embarrassment sort of thing ('hey, is this guy a bit chubby?'), or maybe a commercially sensitive thing ('he's interested in obesity stuff; heh, let's raise the price for him', or 'let's show him some adverts for chips', or 'let's contact his rival and let them know he's interested in his weight'). These latter things are behind the feature where you first search for a flight and the price is great, then the next time you look, it's gone up - allegedly.
  • There's a more fundamental IP issue though -  The simple disclosure of a search term can be commercially damaging, and potentially stop the development of life-saving therapies. The simplest case is chemical structure and drug patents. The most important patent claim in drug discovery is to have composition of matter (and don't get all hissy over pharma misusing the patent system, since patents are absolutely essential for the development of new medicines, the treatment of disease, improvement of food supplies, for funding future R&D, for a source of employment, license revenues to Universities, and taxation revenues, etc). This composition of matter is a claim of a novel chemical structure, that no-one has disclosed before, and it is useful for something. If the structure is not novel, then the patent can be readily invalidated.
Hopefully, you'll understand our reasons for maintaining both user and query privacy.

For an extra clear clarification - we do not, and cannot examine queries of users ourselves within the development team here at the EBI. In case you read the above text as sharing stuff solely with third parties.

Your use of ChEMBL is private, and always will be.


2 comments:

Christopher Southan said...

John, in regard to your second point there is (unless anyone knows otherwise) no patent case law precedent for a successful composition of matter opposition or invalidation based on the interception of chemical (or sequence for that matter) database queries. Strictly speaking the issue is the public exposure thereof in silico. Ipso facto I'm not sure your (or anyone els's) server cache would count as this in court (hacking in would be criminal interception). Until such time as a test case is prosecuted successful the risk remains close to zero, compared to, say, putting your lead structure on a poster.

John Overington said...

You are right. There are some legal defences for 'accidental' disclosure, and also for malicious interception sort of thing. But these seem to go back in spirit to the olden days of real physical post and not electronic transmission.

But I think you misread the post - or a lot more likely I wasn't clear.

The point I was making that you can lose novelty by 'publishing' the query list. The sort of thing I mean is sharing the query list with 'the public', making the query list downloadable, opening up a searchable database of the queries, etc.