Homepage News

Searching for new drugs: relying on the "imagination of computers"

05.05.2017 Health, Technology, Recommended

Photo: Fotolia

A computer equipped with appropriate software is able to see complex relationships that the human mind would never notice. We just have to let the machine find the rules - Dr. Rafał Kurczab, who uses machine learning algorithms to search for new drugs, said in the interview with PAP.

Machine learning is a concept, according to which instead of programming computers to perform specific tasks - the traditional approach - they can be programmed to recognize complex patterns and improve their performance based on previously acquired data.


Machine learning allows the Google search engine to get more and more effective and learn what users need from it. Sometimes even a few letters are enough for it to "guess" what we are searching for. Machine learning is also increasingly useful for banks, retailers, marketing companies - to precisely analyse customers' needs based on their prior activities.




But machine learning is also used by scientists. "for example, I use machine learning algorithms to look for new biologically active compounds" - says Dr Rafał Kurczab from the Institute of Pharmacology pas in Kraków said in the interview with PAP. He added that machine learning algorithms help optimise the selection of reagents in chemical synthesis, which in turn allows for a more effective search for new biologically active compounds.


"With artificial intelligence, we can more quickly and efficiently search large databases - containing information about millions of compounds that have already been synthesized or that can be produced" - said Dr. Kurczab.


He gives an example: his colleagues from the Jagiellonian University asked him to select compounds that could be prepared from commercially available chemical reagents that would most effectively work on a particular biological target. "If we had taken all the reagents that would fit into this reaction, we would have received over 51 million different compounds. We used various models and made a list of the 100 compounds that could be most effective. The colleagues from the Jagiellonian University synthesized them and 80 of them turned out to be active"- described Dr. Kurczab.


The tasks with which algorithms tackle may vary. Sometimes scientists show the algorithm the binding site at the receptor, which has a specific shape. The algorithm searches for a compound that can fit well in this shape. Sometimes, in turn, only the compounds that act on the given receptor are known and researchers have to find similar compounds, even though the structure of the receptor itself is often unknown.


"A drawing of the compound would do for a chemist. He would know how to look for similar compounds. A computer needs to understand what that a compound is" - said the scientist. He described that in search of new compounds, algorithms can use spaces that consist of hundreds of dimensions.




The position of a point in a three-dimensional space is described by three variables. But why stop at values? After all, objects can be described by a multitude of variables, or "dimensions". If, for example, we want to describe chemical compounds, the length of the hydrocarbon chain may be specified on one axis, on another - presence of aromatic ring, on another yet - presence of different elements or functional groups, etc. Mechanical learning algorithms can combine these data and "see" the multidimensional shape, which carries a lot of valuable information.


The task of the algorithm will now be to find similar shapes in the database. "A man, even if he spent a long time on the data, would not be able to analyse them effectively. We simply do not have enough imagination" - said Dr. Kurczab.




"First we >>feed<< the algorithm with input data - describing the compounds that we know are active" - the researcher told PAP. He explained that the computer itself has to find similarities between these compounds and identify the common properties. And then the program searches for objects with similar characteristics.


"Machine learning algorithms are able to perform assigned task in a very abstract way" - said the chemist. He noted that such programs often find objects in the database whose similarity to reference compounds is not obvious at all. And yet in many cases their selection turns out to be spot on.


Dr. Kurczab uses machine learning algorithms to search for new drugs. But he noted that these algorithms, used increasingly in various spheres of life, can be used by researchers from many different fields. "Free software is available online. It can be used by chemists, biologists, psychologists, all those who need to process large and complex data sets in their work" - he said.


PAP - Science and Scholarship in Poland, Ludwika Tomala


lt/ agt/ kap/


tr. RL

Share this
Rating: 0 votes


Forgotten password


Comments: 0
Add comment See all comments  

Most popular sites