AI has the power to enhance decisions about prevention, diagnosis and treatment of health conditions by assisting scientists, health care professionals, and patients.
Machine learning, a subfield of AI, can do what humans cannot, such as processing vast amounts of data rapidly and accurately, and making new predictions based on learning rules and insights from data. This predictive power is especially effective in disease diagnosis and in the discovery of new treatments. For example, machine-learning enhanced drug discovery enabled BenevolentAI to identify two lead compounds for ALS. BenevolentAI's platform produced a ranked list of potential ALS treatments together with biological evidence. Our team were able to rapidly triage these predictions using strategies focused on pathways implicated in multiple ALS processes. BEN-XX1, a compound showing improved CNS exposure showed a profound rescue effect in ALS patient cells. We are currently in late stage lead optimisation of BEN-XX1 to nominate a clinical candidate.
While machine learning is powerful to find patterns in data, human intelligence is not only data-driven, it’s also reasoned and experience based. Today, machine learning technologies can rarely explain the patterns they do find: they are often “black-boxes,” that fail to produce real understanding. This results in a lack of trust in AI tools and therefore poor adoption from doctors and researchers. We need to develop AI that goes beyond simply finding patterns, and helps rationalise its findings so that users can both learn and also help make better AI tools. Our drug discovery scientists routinely propose new methods to enhance the training of our models, whether it’s through modifying the data or new learning mechanisms. Equipped with smart technologies, our scientists can unleash their creativity and explore biology and chemical space in a way that was not possible before.
Providing patients with the ability to take an active part in managing their own health, will result in better outcomes. Through technologies such as natural language processing and text mining, AI could extract relevant, up-to-date information and even summarise medical texts in a way that patients can comprehend. With improved and accessible information, patients are better able to make decisions about their own health, and this will increase compliance with treatments. The use of IoT (Internet of Things) monitoring devices can be used as health companions that make recommendations for a healthy lifestyle. Enhanced with analytical capabilities, these devices can also alert a person when there is a deviation from healthy values, urging them to take preventative measures and consult a professional, or directly alert their care team.Closer health monitoring can result in earlier and better diagnosis of illness. Often, the earlier a disease is detected, the greater the chance of successful treatment. Yet, according to Cancer Research UK, 46% of all cancers diagnosed in 2012 were not detected until stage three or four. Identification of the first symptoms through smarter devices and analytics has the potential to significantly change these statistics.
Since AI technologies are often only as good as the data they rely on, it is crucial to safeguard the privacy, availability and quality of healthcare data.
Comprehensive data protection is essential to ensuring that data is used for the benefit of those from whom it derives. The European General Data Protection Regulation (GDPR) or the Health Insurance Portability and Accountability Act of 1996 (HIPAA) in the US are examples of the broad recognition of this need, as they offer control and transparency to individuals about the use of their personal data. People should have a say in what their data can be used for: in order to facilitate this, we must better educate people on their rights, on the ways in which sharing can benefit and harm them, and give them a chance to affect policy, for example through participation in review boards.
Today, the vast majority of participants in genome-wide association studies (GWAS) are of European descent (Genomics is failing on diversity, Nature 538, 2016). This has serious consequences for medical research, such as understanding disease risk in different populations, or drug effectiveness. Other ethnic groups continue to have disproportionately higher incidence and mortality rates for multiple cancers (Artificial intelligence can entrench disparities – here’s what we must do, The Cancer Letter, 2018). We urgently need to increase data diversity in order to enable more equal treatment opportunities. Recent initiatives, such as the establishment of the first pan-Africa biobank, are starting to change the landscape through purposeful funding efforts.
Biomedical data fragmentation is a major problem that inhibits progress. Patient data is often both limited and siloed, whether it’s in the pharmaceutical industry or in hospitals. In addition, data that could enhance drug discovery is also hard to come by, especially negative data. In recent years, the increase in large data consortia uniting government entities, hospitals, charities and pharma companies is a step towards democratizing data access to benefit research and development.
The future of healthcare looks brighter with AI, because it can enable higher prevention, more accurate diagnosis and more novel treatment options for our growing patient population. There is a real eagerness and need to build on synergies between stakeholders and experts in healthcare. In order to do so, certain barriers need to come down (such as data fragmentation) and others need to be put up (such as data protection). This is a process that should involve everyone: medical institutions, pharma and tech industries, governments, and patients, working together.Alix Lacoste, VP Data Science
VP Data science