In a previous post I described how to use cheminformatics to determine whether or not a compound was radioactive by checking for unstable atomic nuclei. Alas, such a binary yes-or-no classification is a clumsy tool for eliminating dubious structures, or even identifying non-drugs in chemical databases. For example, bismuth is unfortunate enough to have no perfectly stable isotopes, making drugs such as GSK’s Pylorid (ranitidine bismuth citrate) technically but negligibly radioactive, even though bismuth’s half-life is 1.9×1019 years, or over a billion times the life of the universe. Likewise, although [7H] has been observed experimentally, its half-life of 23 yoctoseconds (1×10-24 seconds) makes it use impractical for traditional drug discovery.
As an interesting aside, measurements of half-lives form an interesting exception to usual SI units, with small values being in fractions of a second (milliseconds, microseconds, nanoseconds), intermediates values being in minutes, hours and days, and large values being in multiples of years (kiloyears, megayears, gigayears and so on). Converting units to normalized form, for example times in terms of seconds, is all part of scientific computing.
To refine the filtering of plausible/reasonable neutron counts we propose the use of molecular half-life, rather than binary categories such radioactive vs. stable or experimentally observed vs. purely hypothetical. The one subtlety with this definition is that a molecule’s half-life is determined from the half-lives of all of the unstable atoms it contains. As hinted in the title, a molecule with two copies of the same unstable nuclide, has a half-life half that of an individual atom. In general, the formula for a molecule’s half-life is 1/Σi{1/t½(i)}, or the reciprocal of sum of the reciprocals of constituent atomic half-lives.
At NextMove Software, we currently use the nuclide half-lives tabulated by the Nubase2003 database (downloadable as an ASCII file). The resulting “half-life validity” check can be used to identify dubious structures in chemical databases using a suitable threshold. For example, the most suspicious isotopic specification in NCBI’s PubChem database belongs to CID 11635947. This is a structure deposited by NextBio that contains an erroneous [8C]. Although [8C] has been experimentally observed (and PubChem should be congratulated for containing no nuclides that haven’t been observed), it has an impressively low half-life of only two zeptoseconds (2×10-19 seconds).
A more reasonable threshold might be around the 1223 second (~20 minute) half-life of [11C], which legitimately appears as the least stable compound in Accelrys’ MDDR database. 11C is used as a radiotracer in Positron Emission Tomography (PET), where compounds have to be used within about three half-lives of their manufacture. When filtering compound screening collections, threshold half-lives much higher might be reasonable.
My final observation is that even more accurate calculations of molecular half-life is possible by taking into account the influence of chemical environment on atomic half-life.
For example, metallic Berylium-7, [7Be], has a different half-live to covalently bound Berylium-7, such as in Berylium-7 fluoride, F[7Be]F, or Berylium-7 oxide, [7Be]=O, and fully ionizied Rhenium-187, [187Re+75] has a half-life of 33 years, significantly lower than that of metallic Rhenium-187, [187Re], which has a half-life of 41 gigayears (41×109 years).
Image credit: Ed Siasoco (aka SC Fiasco) on Flickr