Artificial Intelligence

Does Artificial Intelligence Design New Drugs Or Discover Them?


What is discovery? In mathematics, there is this age-old question of whether new math is discovered or invented. It makes sense to ask the same sort of question about modern drug discovery. When using artificial intelligence to identify drug candidates, are these new drug candidates being developed, or simply exposed through a process of narrowing down the possibilities using mathematics and science? Are these new drug candidates discovered or designed? Or maybe it’s a distinction without a difference.

A flurry of progress in the race to identify a COVID-19 vaccine has produced new automated techniques for drug discovery using artificial intelligence. For example, artificial intelligence is being applied at drug companies such as Benevolent AI, to find drug candidates among existing medications. In terms of therapeutics, Benevolent AI uncovered 6 molecules that entered the clinical validation process. Beyond discovering new drugs, companies such as Innoplexus, Deargen, Gero, Cyclica, Healx, VantAI and others are using artificial intelligence to repurpose existing drugs for new uses. Companies such as Insilico Medicine, Exscientia, SRI International, Iktos and others are inventing new drugs altogether utilizing AI. Biopharmaceutical companies across the world are already adopting AI strategies to integrate their discovery process. For example, Atomwise is accelerating the molecule discovery process for pediatric cancers by implementing deep learning algorithms and elastic supercomputer platforms to predict potential medicines. Their primary goal is to reduce the time taken to identify and develop viable therapeutics. Another example is the MELLODY project, which is a blockchain-based solution that aims to develop a machine learning platform that can learn from thousands of sets of proprietary data generated during the drug discovery process. The project, once fully developed, could make it easier to identify which small molecules show the most promise for future research. These companies are part of a broad-based shift of more and more companies applying machine learning to find new uses for drugs, and new drugs to try out. When algorithms can do drug discovery, it seems reasonable to ask: “What does drug discovery really mean?”

Insights Hidden In Plain Sight?

To answer this question, it helps to look a bit more closely at applications of artificial intelligence in drug discovery and related fields, to get a sense of the practical and business implications of this new technology. Artificial intelligence has a wide array of applications in drug discovery, and research labs have translated their progress into the commercial stage based upon early work with these techniques, exploring the intersection of drug development and algorithms that can learn. Even NVIDIA got in on the action. One approach is computational, where many combinations of molecules are “tested” in the computer. Another approach is discovering insights from latent information, and is popular well beyond the drug discovery arena. Big data analytics has evolved a new and exciting set of techniques for distilling new insights from existing data. For example, an exciting approach was recently published in Nature that assessed how different chemical properties of materials are related within the text of a big dataset of research papers. In that work by researchers from UC Berkeley and Lawrence Berkeley National Laboratory, rather than looking at the molecular data on these materials, found out that using unsupervised learning to expose materials science knowledge present in the scientific literature could be used to recommend materials for functional applications several years before their discovery. This idea that new knowledge is hidden in plain sight is pretty interesting for drug discovery. In 2017, a researcher from Chuo University in Tokyo published another compelling technique in Nature, that identified a combined set of genes and compounds that significantly overlap with gene and drug interactions. The researcher used this technique to identify two promising therapeutic-target genes, and for their protein products, they identified a promising candidate drug for cirrhosis (a common illness with few good treatment options).

Big Data

According to a recent Deloitte report, the use cases of artificial intelligence in the drug discovery industry could speed up and reduce the cost of the drug discovery cycle. Usually, this cycle takes five to six years from the beginning of the research and discovery phase to the preclinical testing. It averages ten to twelve years and roughly $2 billion per drug, just to launch and productize. Yet, according to Deloitte, when this new drug reaches the market, the expected return on investment is under 2%. The drug industry would be more profitable if discoveries were more automated, reducing the cost of launching new medicines into the market. Artificial intelligence is a promising solution to support new drugs’ early development. AI solutions can decrease the time spent in the drug discovery phases and speed up its most critical aspects: the discovery itself and the preclinical stages by a factor of 15, according to the report from Deloitte. This sort of project requires a lot of data to mine, and definitely requires fine-tuning that makes the process seem more like science and discovery than engineering and design. Maybe the truth is really somewhere in the middle.

Discovery Or Design?

Looking back at the adoption of software into large scale project like the human genome project, the process of drug discovery, and the pace at which companies are utilizing AI for drug discovery and research, would not be a long shot to say that by 2030, drug discovery processes will mostly be driven by AI software. Timings from screening to preclinical will be reduced significantly and new drugs capable of treating extremely specific pathologies will become the new normal.

So the big question here is whether artificial intelligence approaches really discover new drugs, or simply designa them through a process of optimization? It is really the former (i.e., discover) and not the latter (i.e., design). Artificial intelligence for drug discovery is all about getting lucky. These approaches are optimizing, but they need a lot of luck to reach a “right” solution. Using every screwdriver in the toolbox until one of them fits the screw is not the same thing as inventing a new screwdriver. Now we ask the question again: is this invention or discovery? In this case, we can see the inside of the machine that is producing new drug candidates, and indeed it is learning from data, and optimizing to select options that fit a pattern. Discovery it is. And this field is only just getting started. The future is bright.



READ SOURCE

This website uses cookies. By continuing to use this site, you accept our use of cookies.