We’re only two months into 2023, but already, one of the big stories of the year is shaping up to be the coming-of-age of AI. Last year, tools like DALL-E showed off how AI could generate complex and dazzling images. Now, with ChatGPT out and the incorporation of AI into Microsoft’s search engine Bing, the disruptive potential of generative AI across multiple industries is increasingly clear.
But, how can AI be applied to improve human health? There are well-justified concerns about AI’s potential for bias and causing unintended harm, but I am hopeful that artificial intelligence can also be harnessed to help people live longer, healthier lives. Here are what I see as three best-use cases for AI: in basic research, in patient care, and in defining disease.
Using AI to understand basic biology
Modern biology has been transformed by big data, with genomics, proteomics, microscopy, and robotics making it possible to measure biology’s complexity with increasing breadth and precision. But making sense of this big data can be challenging, which is where machine learning comes in.
Take Leaps portfolio company Gandeeva Therapeutics. Gandeeva focuses on understanding proteins, the molecular machines inside cells. AI-based tools like AlphaFold and RoseTTAFold have made enormous strides in predicting protein structure, but it’s still incredibly challenging to predict how small-molecule and biologic drugs will bind to proteins and alter their functions. Gandeeva is combining AI with novel cryoEM imaging to bridge the gap between understanding protein structure and designing new drugs. By taking very high-resolution cryoEM images and using them to refine their molecular models, Gandeeva hopes to build a positive-feedback loop that will help us better predict protein-drug interactions and accelerate drug discovery. In fact, Gandeeva’s proprietary platform successfully created the first images of the SARS-CoV-2 Omnicron spike protein, a crucial step in helping in the fight against the pandemic.
Another example comes from Recursion, which is building a high-throughput phenotyping platform. Cellular phenotypes are essentially biological states, and understanding how hundreds of biological components interact to produce these phenotypes is one of the central questions of modern biology. Recursion has built an integrated hard- and software platform that uses robotics to perform high-throughput measurements, then leverages AI to infer and predict biologically relevant molecules and the relationships between them.
Both Recursion and Gandeeva embody the thesis that high-throughput data collection and AI-driven advances in data science can be integrated to advance our understanding of biology. It’s like building a map — you need the right measurements and the right framework to know where you’re going.
Leaps portfolio company Dewpoint Therapeutics embodies yet another way in which AI can advance our understanding of basic biology. Dewpoint focuses on understanding and developing drugs for diseases caused by problems with cellular condensates or proteins and nucleic acids. Previously, cellular condensation was difficult to track because we couldn’t separate the signal from the noise. But using AI, we can home in on this phenomenon and understand how it contributes to disease. As Dewpoint, Gandeeva, and Recursion show, partnering with AI can be a powerful way to enhance our understanding of the basic biology that underlies disease.
Another avenue through which AI holds great potential is by transforming our definitions of human disease. Most diseases have traditionally been diagnosed by their symptoms, even though different biological drivers may be responsible for these effects. Cancers, for instance, have traditionally been classified by the tissues in which they arise, but it is becoming clear that treatment success and failure may depend less on where a tumor is located and more on the tumor’s molecular drivers. Likewise, genetic variation and other molecular drivers can contribute to chronic diseases like cardiovascular disease, with varying implications for patient treatment. Tailoring patient treatments to these underlying biological drivers is a major challenge for medicine in the 21st century.
Using AI to design drugs
Here, too, I am hopeful that AI can be a powerful partner. Machine learning can help us make sense of vast troves of data, including population genetics and real-world clinical data, to guide disease diagnosis and develop more personalized treatments. In addition to Gandeeva, another example comes from Leaps investment Cellino. Cellino’s goal is to scale up the manufacturing process for autologous cell therapies, which are traditionally time-consuming and difficult to produce from patient cells. Using AI and innovative laser technologies to standardize cell manufacturing, Cellino hopes to make these tailored treatments available and affordable for patients worldwide.
At Ukko, another Leaps portfolio company, scientists are using computational tools to reengineer allergen proteins and then create medicines to help patients with severe allergies become more tolerant to foods such as peanut or other common, severe allergens. The goal of the lead program for peanut allergies is to develop a tolerizing vaccine that exposes patients’ immune systems to the computationally-reengineered peanut protein, but without the risk of anaphylaxis. We hope this will dramatically expand the safety, efficacy, and accessibility of allergy tolerization treatment, and so I’m excited that Ukko’s peanut vaccine is moving quickly toward clinical studies.
Using AI to enhance patient care
AI has the potential to expand basic research and discovery, but what about patient care? ChatGPT has shown that AI-powered chatbots can gather information, answer questions, and simulate conversations. Portfolio company Woebot Health has built an AI-powered chatbot that uses natural language processing to interact with patients and serve as a mental health resource, based on a clinically validated model. Woebot is part of Leaps’ commitment to transform health with data, with rigorous clinical studies to back its approach. Ada Health, meanwhile, allows patients to assess their symptoms, built on AI and clinical evidence. These services are available around the clock, and they can provide real-time information and feedback to patients and encourage them to seek more advanced care when they need it.
Another approach to digital health comes from Huma, which develops remote monitoring platforms and wearable devices to enhance patient health. Recently, Huma has partnered with Bayer to develop an online risk-assessment tool for cardiovascular health. Huma’s algorithm helps patients assess their heart risk factors from the comfort of their own homes and share their results with healthcare professionals as part of long-term care. This partnership aims to reach more than 100 million people across the U.S. who may be at risk of cardiovascular disease.
As with any new digital technology, it will be crucial to design strong safeguards for patient privacy to promote mutual trust between patients and providers. Once these safeguards are in place, I’m hopeful that these digital, AI-guided platforms will routinely enhance traditional healthcare.
It’s clear that machine learning and big data are powering the life-sciences revolution, and I am optimistic that AI can help us make sense of the complexity of biology and enhance patient care. I’ve highlighted several companies that use AI as a core feature of their platforms, but many of our investments are also harnessing this technology in different areas, such as to improve agricultural practices. I look forward to seeing all the many ways we can leverage these powerful technologies to help people around the world live happier, healthier lives.