You are here

Alternative methods

14 - 02 - 2019

Can computer models evaluate the toxicity of chemicals?

Can computer science replace animal models to evaluate the toxicity of a substance? This is the question to which an article in the journal Nature, published on January 14, attempts to answer.


The demands of the authorities in Europe and the USA on the knowledge of the toxicity of marketed chemicals have increased. In the US, the number of vertebrates used for this purpose has increased from 6,500 in 2016 to 76,000 in 2017. The same authorities have committed to reduce the number of animals used for scientific purposes, so other methods are being looked into.

In this field, specific computer programs use data obtained from a large number of substances tested in animals and in vitro to estimate the toxicological profile of new substances by comparing their chemical structures.

The recent access to powerful computer resources and large databases (140 million substances) such as that of the European agency ECHA has consolidated this approach. Of the 190,000 substances tested, the accuracy of a program's predictivity has recently been 87%.

Other researchers are combining different in silico and in vitro models to understand the mechanisms linked to toxic effects and thus predict acute toxicity, endocrine disrupting effects, or the effect on embryonic development of an unknown substance.

Today the authorities are investing in these works but have not yet decided to completely work without animal models. Indeed the value of computer modeling depends on the knowledge we have of what we want to model, aka life. But we are still far from knowing how a living organism works, and therefore modelling it reliably.

However, with the upstreaming research, it is possible to use all the accumulated and recorded knowledge to eliminate the substances likely to be highly toxic, and in some cases, replace animal models or at least reduce them. The work is still ongoing.