AI and the Lab of the Future
One sunny day in the future, a person will walk into a clinic and their wearable device will download their health record. A quick blood test or body-scan and the clinician will have a prognostic profile and risk scores to determine what if any follow up is required. If there’s something wrong, a precise test can be administered to point toward an optimal therapy. Where no optimal therapy exists, a tailored one will (rather figuratively) fall from the sky, designed by an AI that has learned from the biology and outcomes of all the patients that came before.
The laboratory of the future does not look at all like what we think of today—a fluorescent-lit room bound by four walls and a fume hood. The laboratory of the future will be a network of physical and virtual spaces, spanning the entire lifecycle of therapeutic products. Robot-guided molecular and cellular biology labs test predictions of new chemicals generated in the cloud. Biomarkers derived from organs on a chip and federated global clinical trials yield patient-precise clinician reports. Wrist watches and smartphones monitor compliance and collect troves of data over time. In fact, all of these technologies are here today, but the laboratory of the future will learn to connect them into constellations of discovery.
Artificial intelligence (AI) enables each of these laboratory pursuits. Machine learning (a subset of AI) algorithms detect patterns hidden in vast data, output predictions that leapfrog conventional experimentation and form the backbone of automation and decision support. But more crucial to this vision of the future, AI-enabled technology can form the connective tissue that creates virtuous data feedback loops. The resulting advances in medical innovation will not be sequential and phase-gated, but rather concurrent and transformational.
Already, AI-based solutions are being deployed to overcome challenges and bottlenecks up and down the spectrum of healthcare and life sciences. The promise and hype, but also progress and early wins, are perhaps most palpable in the drug discovery space. Innovators and entrepreneurs and industry veterans alike are embracing the potential of machine learning—they share the common goal to break down historically insurmountable barriers to bringing new and more efficacious therapies to patients in need.
The laboratories of the future are being dreamed up, and built, to foster greater automation and autonomy, near real time feedback, intelligent process controls, experimental and clinical decision support and, ultimately, tailored solutions that yield the best care for patients. But if this spectacular range of application areas defines the “laboratory of the future,” we must clarify— which laboratory?
One can find noteworthy examples of AI-enabled disruption in various laboratory settings, including: high-throughput drug screening via cell-based imaging (e.g. Recursion Pharma); generative chemistry to imagine all new medicines (e.g. Insilico Medicine); automation of robotics and experimental execution (e.g. Arctoris); preclinical modeling and new biology discovery (e.g. Insitro); translational and clinical biomarkers (e.g. Genialis); in silico clinical trials (e.g. Unlearn.ai); and clinical decision support at the point of patient care (e.g. Molecular Health). Some teams are going big to integrate AI at each stage of the process (e.g. Valo Health), aiming to bend the old linear chevron diagram of drug discovery and development into an entirely closed loop, with tight feedback at each step.
The lab of the future looks spectacular—an updated version of the Jetsons, or maybe something like Tony Stark’s workshop, but for medicines and diagnostics. But as anyone who has worked at a real lab bench can tell you, the reality has less to do with blue mood lighting and techno beats and more to do with painstaking trial and error. The same perspiration quotient holds true for data scientists, for whom the vast majority of effort goes into collecting and cleaning data. And thus the lab of the future will require this grunt work, with a focus on quality control, independent validation, and anti-bias measures.
The AIs that orchestrate the lab of the future will need to learn to do their jobs. And learn they will, based on whatever training data are fed into the system. The outputs will be determined by the inputs (as in the truism, “garbage in, garbage out”). But even if we take care to avoid garbage, the inputs still may not be representative of the real world in which the AI needs to function. Data usually are biased, reflective of their origin story: patient demographics, geographic constraints, experimental idiosyncrasies, collection methods, processing technologies, self-selection of volunteers, economic barriers to participation, etc.
Researchers building the lab of the future are keenly aware of these sources of bias. Methods exist to identify bias, and even to circumvent certain biases with statistics and corrective measures. But biases that are built into the very fabric of our healthcare system, and therefore directly influence the composition of datasets, will be much harder to remedy. Thus the very success of the lab of the future depends on a shared responsibility to make healthcare more equitable and participation maximally ethical, to address tensions inherent between data rights and data access.
About the author
Rafael Rosengarten, Ph.D., is the CEO of Genialis, and leads Genialis’ effort to integrate and mine vast and diverse sources of biomedical knowledge to realize the promise of precision medicine and therapeutic discovery. He spent nearly 20 years in biomedical research prior to Genialis, publishing on the evolution of innate immune systems, bioengineering of microbes, and genetics of development. Rafael co-invented the j5 DNA assembly design automation software for high-throughput molecular design and analyses (since commercialized by TeselaGen). Rafael earned his doctorate at Yale University, and conducted postdoctoral research at Lawrence Berkeley National Laboratory’s Joint BioEnergy Institute (JBEI) and Baylor College of Medicine. In his free time, Rafael enjoys cooking and rock climbing, and raising heirloom tomatoes and two precocious children. He can be reached at firstname.lastname@example.org.
This article was also published on Bio-ITWorld.com in May 2021.