The future of Artificial Intelligence in precision medicine

Pedro Amorim

Phrases like “Genomics and Artificial Intelligence, the next Holy Grail of Precision Medicine” are seen much more frequently in recent texts and publications. Projections of the impact of Artificial Intelligence (AI) systems on the global healthcare market point to a strong expansion in the coming years. A boom of new companies that converge IT and bio-computing is fueling yet another market competition between Chinese and Americans.

But, after all, in such an optimistic scenario for the impact of AI on the future of precision medicine, what remains for us to wait for the great discoveries?

AI meets genomics

In research published by consultancy Frost & Sullivan [1], AI-based systems are estimated will generate $6.7B of global healthcare revenue by 2021, a significant increase compared to the $633.8M generated up to 2014.

The great barriers to the development of precision medicine, costs and technology, have been overcome. One of the AI ​​techniques, the Machine Learning, has a prominent role in this scenario. This technique, broadly speaking, seeks to identify patterns in data sets, and can, for example, be applied in the development of predictive models for an individual’s tendency to develop a disease or respond to treatment [2].

More than that, AI has been seen by many as an important research tool in genetics. Which makes a lot of sense if we think that, despite the increase achieved recently of the sequencing and even editing of genes, we still know low about the functions of the human genome and how they reflect on our health and physiology.

Jun Wang, leader of the Chinese multinational company BGI, said in an interview with Nature magazine: “Life is digital, like a computer program – if you want to understand the results of programming, how genes lead to phenotypes, you need an AI system to discover the rules”. He brings a vision that places AI as one of the main resources for new discoveries in genomics, a field of genetics that studies all the genes of an organism and how they are organized. But one of the first questions still to be made public is: What types of progress can the AI ​​and genomics combination make in the coming years?

The next few years

Many public and private research projects have been highlighted in the search for new knowledge of the origins of our phenotypes. A search for personalized health insights. These projects have created large masses of data, genomic, physiological, health, environmental and lifestyle data, which are used for the development of AI algorithms. It is worth highlighting the project of BGI and Jun Wang, previously mentioned.

In addition, the future of preventive medicine must also be impacted. The importance of early detection of diseases is indisputable, either for better control or to increase the chances of cure. Now, imagine the possibility of detecting genetic markers responsible for cancer in a simple blood test. This is one of the big ideas today behind liquid biopsy and part of its viability is technology for analysis. Having the ability to find genetic markers at population and personal levels with medical and statistical rigor will require techniques such as Machine Learning and its power of comparison on a large scale [4].

Last, but not least, are the research into new therapeutic options for cancer. Renowned institutes with the Mayo Clinic, as well as multinational companies like IBM and Pfizer [5], have established major projects involving AI for the investigation of targeted genetic markers in personalized cancer treatments.

These major research movements, despite the uncertainties and risks they involve, generate high expectations in the healthcare ecosystem as a whole. But are the contributions of AI and genomics convergence all medium/long term? Can clinical practice no longer benefit and bring great learning to this future of precision medicine?

And today?

Well, when one of the AI ​​gurus, Andrew Ng, stated that it would be the new electricity, he wanted to emphasize the vast field of applications it would have. Even within precision medicine, this vision also opens us to several possibilities of impact.

The analysis of the genetic variants, the result of a Next-Generation Sequencing (NGS), for example, is a very analytical and often costly process in the search for relevant information about the variants in order to identify the one responsible for the clinical phenotype.

There we have great opportunities to contribute AI-based algorithms. Scores of pathogenicity and damage prediction of the alteration, identification of artifacts, ranking or even the classification of the variant itself are examples of tasks prone to the development of Machine Learning models of classification and ranking.

In another case, the search for relevant content in publications can be optimized through models of Natural Language Processing (NLP). These are just some examples of the impact that this development can bring to the analysis process, whether by presenting new information, by the agility and precision in the process of searching for content, by prioritizing the variants to be interpreted, among others. And as a feedback system, consolidated improvements in the analysis generate new data, always with more quality.

Of course, not everything is just flowers and the development of these models requires great technical-scientific rigor and still presents classic challenges such as validation and demanding extremely high sensitivity of the models. I can also say, from my experience with other Machine Learning projects in health, those of variant classification are those that involve the greatest complexities of building datasets and validation since the bases can contain several sources of bias. These challenges, however, are now overcome by accessing good databases and good chemistry among technical and clinical experts during modeling.

In short, we have a super common scenario in the development of AI in other sectors: large medium/long-term projects led by large players that solve broader problems (Ex: Autonomous Cars); as smaller players, today symbolized in the figure of startups, developing niche solutions, with expert algorithms and with a lot of impact on current practice. In AI, these groups are divided into Horizontal AI and Vertical AI, respectively [6].

Well, this was a small illustration of what I have seen and been able to participate, through Varstation, in innovation in Genomics through AI. It was super important for us to understand our role and start quickly to bring our contribution to the ecosystem. Varstation today has several AI-based solutions that have made it possible for us to empower analysis and democratize genetics as a whole.

References: