An AI-based program managed to design a new drug in 46 days instead of the eight years it takes on average for 'human' researchers. The work of Insilico Medicine start-up and the University of Toronto, which have been looking for a potential cure for fibrosis, that is, the healing of tissues that occurs in certain diseases, is described in Nature Biotechnology.
The algorithm, whose code was made available to everyone, examined all previous research on molecules that targeted a particular protein essential in the fibrosis process, giving priority to new structures that could be synthesized in the laboratory. In 21 days, the program created 30 'candidates', six of which were actually synthesized. Two of these were tested in cells and the more promising of the two also in mice, showing 'drug-like' activity against the protein.
The whole process took 46 days and about $ 150 worth of funds, much lower than traditional methods. “This study - conclude the authors - illustrates the usefulness of our models for a rapid design of molecules that are easy to synthesize, active against a specific target and potentially innovative”.
How much can artificial intelligence help us create more effective drugs and, if we want to widen the horizon, to improve medical therapies (therefore, our health)? Much, according to the experts who will participate, on Wednesday 9 October, at the Forum "How Artificial Intelligence can change the pharmaceutical landscape".
The Forum will then be followed at the 18.30 by the Convivium, in Italian, on the theme "Future + human", during which Boas Erez, rector of the University of Italian Switzerland, and Alessandro Curioni will talk with Francesco Morace, sociologist and president of the Future Concept Lab, on the new scenarios that open up following the ever-increasing use of artificial intelligence and other advanced technologies, and also on related ethical problems (starting from the protection of personal data, which are used in abundance by super-computers).
"Artificial intelligence - says Morace - it will never be able to replicate the human one, endowed with empathy, intuitive capacity and other characteristics that the machines cannot copy. In short, the future will be more and more human, but artificial intelligence will make us discover it". Damian Realini, journalist of the RSI, will conduct the evening.
AI in medicine: from research of new molecules, to diagnostic support
The possible applications of artificial intelligence to the various sectors of scientific research and, more generally, of health are vast: first of all, AI systems allow us to study in a much faster time, and with greater effectiveness than traditional methods, the molecules that appear to be potentially 'active' to cure even serious diseases such as tumors (research laboratories and pharmaceutical companies increasingly use these systems). But artificial intelligence also helps to reposition (so it is called in technical terms) a series of drugs that had been approved for a certain type of disease and instead are able to treat other diseases as well. Thanks to artificial intelligence it is then starting to find remedies for rare and 'abandoned' diseases (or, as they are defined, orphans).
But that's not enough: systems like "Watson for Oncology ", developed by IBM in collaboration with the Memorial Sloan-Kettering Cancer Center in New York (one of the most important cancer hospitals in the world), help doctors to choose the best therapies in the most difficult cases, or to refer patients to "Clinical trials" (the trials of new drugs) most suitable in the world.
IBM is also running a study project together with the Zurich University Hospital to perfect an automatic system that can examine the "slides" (ie tissue fragments taken with a biopsy, or during surgery) and identify precisely the alterations caused by the different diseases, alongside the work of the pathologists. But other companies and research institutes are also moving in this direction.
To function well and to provide reliable results, artificial intelligence systems need huge databases from which to derive information, which are then processed by very powerful computers, 'governed' by ad hoc algorithms and neural networks (ie, systems that in some ways mimic the organization of human nerve cells): this is the "Deep learning".
Ticino is very active in this sector, with IDSIA at the forefront. In this regard, a collaboration agreement has recently been defined between IDSIA and the Cantonal Hospital Authority (EOC, which manages public health in Ticino) for the application of advanced artificial intelligence methodologies to the data provided by the organization.
But other projects are also underway: “One of the most important involves, in addition to IDSIA, also the Tropical Diseases Institute of Basel, the Zurich Polytechnic and the Department of Pharmacology of the University of Geneva - he explains Andrea Danani, head of the computational biophysics laboratory of IDSIA, and scientific coordinator of the October 9 Forum - In particular, we are examining the mechanism of action of an African plant that is active against Chagas disease, which is widespread in Central and South America. In these cases artificial intelligence can provide decisive help".
It is much more difficult to design a new molecule from scratch (a molecule, that is, that does not exist in nature). "So far it has not been possible - says Ed Griffen - because we still do not understand sufficiently accurately the chemical and biological mechanisms that would lead the new compounds to bind to enzymes and cellular receptors, and also the ways in which these new molecules would be absorbed or expelled by the body".
But studies on this side continue, and the prospects of creating completely new drugs through massive use of artificial intelligence appear concrete. It's just a matter of time. Of course, the results obtained by artificial intelligence systems must then be confirmed in the laboratory, using traditional techniques, and it is now clear to everyone that AI systems must work alongside men, help them quickly perform calculations or "comparisons" that would require a lifetime whole, but without ever completely replacing human beings.
"These systems still do not develop thought' - confirmation Boas Erez - but they have a powerful ability to statistically analyze the available data (huge masses of data), improving their performance as they go along. The machines are programmed to learn, following the algorithms written by humans. As with animals domesticated by men, which are pushed to move forward thanks to a series of incentives (the sugars ...), the algorithms also contain 'rewards' for the machines they learn by themselves, in order to stimulate them to do always better".