Toxicity
Such criticisms may be well founded but it may also be argued that it doesn't make economic sense to ignore toxicological data, often obtained at great expense, when many otherwise suitable drug candidates fail at some toxicity hurdle. The early prediction of likely undesirable effects, even in a qualitative fashion, can only improve the efficiency of the search for new therapeutic agents. Social and economic pressures are encouraging the more widespread use of in vitro techniques for toxicity testing. These inherently more simple systems may be expected to produce data which are dependent on better defined interactions than in vivo tests, although the choice of appropriate analytical methods should allow the useful analysis of toxicity data from any source. Modern techniques of molecular modelling allow the facile calculation of a wide variety of molecular descriptors for diverse structures and thus many of the earlier objections to toxicity prediction can be answered, at least in part.
New technologies have been applied to automate the drug discovery process including high throughput screening (HTS) and combinatorial chemistry. Unfortunately, the overheads of running HTS and supplying these assays with sufficient compounds means that the cost of each project can be considerable. Avoiding compounds that cause problems in HTS is paramount to improve efficiencies. Computational filters can be applied to lists of compounds to remove those with toxic functional groups, unsuitable physicochemical properties or chromophores. Other in silico methods that have been applied to compound lists include those attempting to predict biopharmaceutical properties such as absorption, distribution, metabolism, excretion and toxicity (ADMET).
Thus, an application that was viewed as "alarming" in 1978 has now become an accepted and routine part of the molecular design process. Commercial systems are available for toxicity prediction (see links) and many companies are building models of toxicity and other biopharmaceutical properties using in-house data.
A variety of techniques are employed in the prediction of biological activity, e.g. Quantitative Structure-Activity Relationships (QSAR),Structure-Activity Relationships (SAR), molecular modelling and expert systems, and it was perhaps inevitable that these methods would be applied to attempts to estimate the toxic effects of compounds. The first reports of such applications appeared in the early 1970's but received a mixed response, in fact the IUPAC commission on Medicinal Chemistry in 1978 viewed the development of an early toxicity estimation system as " alarming enough" to request a report from the IUPAC working party on QSAR. The objections in this and other reports included the fact that many toxicological data sets consist of non-congeneric compounds, that toxicological phenomena, especially oral LD50, do not constitute well defined interactions, and that death is a non-equilibrium process! The basis of these objections concerns the difficulty of measuring or calculating physicochemical descriptors for diverse chemical structures, the supposition that successful prediction of activity requires a single underlying mechanism and the assumption that there is some rate determining equilibrium process controlling biological effect.
