Trendspotting: What’s Coming for Bio-IT in 2022
January 10, 2022 | We spoke with several players in the Bio-IT World vendor community to gain insights and predictions for the coming year. Once again, despite vaccine availability, the COVID-19 pandemic has heavily colored and steered the conversation. The topics of the day continue to highlight AI and automation. Everyone seems to recognize that the pandemic has exposed inefficiencies and imbalances, and many hold hope that AI can help overcome issues such as healthcare burnout. “The only way to avoid further burnout for healthcare staff is to adopt AI and tech that reduces friction and opens up clinicians’ time for meaningful care,” said Susan Collins at Twilio.
Varun Ganapathi of AKASA predicts increased human-guided AI to improve the quality of algorithms. “Just like people, AI needs to constantly be learning to improve performance.” “Ultimately, people will have machines report to them. Humans will be the managers of staff (both other humans and AIs) that will need to be taught and trained to be able to do the tasks they’re needed to do.”
Predictive AI will see its day. Mark Day of iRhythm tells us that “Health wearables must become both proficient and validated in determining who needs preventive care before symptoms and associated outcome risks manifest”. Ed Ikeguchi of AiCure also spoke of sensor data accelerating precision medicine. “Rather than relying on subjective data such as patient-reported outcomes or one-off check-ins, video and audio-based digital biomarkers can catch subtle cues of a patient’s disease progression and how a treatment impacts their daily life,” he said.
We heard more than once that AI algorithms will increasingly address bias. “Methods will develop to provide greater insight into the “black box” of AI algorithm decisions, which will guide understanding into whether these decisions represent bias based on factors including race, gender, and age,” said iRhythm’s Day. “I expect individuals and organizations to continue discovering errors, omissions, and blunders in their data where biases in collection and storage led to incorrect, misleading, and harmful outcomes,” said Andrew Kasarskis of Sema4. “2022 will be the year of bold action to address social determinants of health, and clinical AI will play a starring role in both understanding how these barriers drive disparities and directing the right resources to the right patients and populations,” said John Frownfelter of Jvion.
Another repeated idea is the need for integrating varied datasets and using multi-modal AI to drive analyses. “Bringing disparate high-resolution data (e.g., multi-omic, imaging, sensors, labs, and clinical) together in a compliant manner unlocks new insights around patient response, biomarkers, safety, and dosing,” said Sastry Chilukuri of Medidata.
Kimberly Powell of NVIDIA predicts that the explosion of known protein structures from projects such as AlphaFold and RoseTTAFold, in combination with AI-generated chemical compounds, will increase “the opportunity to discover drugs by a million times.”
Another change for 2022 may be more certain, as it involves regulatory oversight. Josh Gluck of Pure Storage says requirements under the healthcare interoperability rules will help standardize patient records and reveal factors influencing patient outcomes. “Payers will have the ability to dig deeper into some of the social determinants of health and chronic disease that can open doors to greater patient engagement. Real-time analytics will be essential to this vision, though, and can only be achieved if the right foundation is in place,” he says.
Here are the full trends and predictions, including additional forecasts for newer ‘omics, green chemistry, SaaS medical devices, federated learning, XNAT, and protein sequencing. –the Editors
Andrew Kasarskis, Chief Data Officer at Sema4
ML and AI Momentum: We continue to see great advances in machine learning (ML) and artificial intelligence (AI) applied to large information-rich data sources in fields such as image analysis and natural language processing, and I don’t expect that will slow down at all. Some of these algorithms are already being successfully applied to biomedical data and are great at grouping and classifying data and entities represented by vectors, matrices, or cubes of data.
Efficient allocation of data curation resources: This is a need for technological and process innovation that I’d love to exist but don’t yet see happening. When obtaining those large corpuses of well-labeled data to train the AI, some human manual and semi-manual work is inevitably needed. This work is always expensive, never scales well, and frequently takes experts with esoteric knowledge away from important value-generating activities. Figuring out the most efficient way to allocate manual curation work seems, to me, like a significant unmet need that impedes progress in the use of data technology, particularly in biomedicine.
Continued focus on data equity: Societal biases and inequities can be present whenever data is used. I expect individuals and organizations to continue discovering errors, omissions, and blunders in their data where biases in collection and storage led to incorrect, misleading, and harmful outcomes. Continued focus on identifying and resolving these issues is important for both accuracy of conclusions and equity in data use.”
Kimberly Powell, Vice President of Healthcare, NVIDIA
AI Will Generate Million X Drug Discovery: Simultaneous breakthroughs of AlphaFold and RoseTTAFold, creating a thousand-fold explosion of known protein structures, and AI that can generate a thousand more potential chemical compounds has increased the opportunity to discover drugs by a million times. Molecular simulations help to model target and drug interactions completely in silico. To keep up with the million times opportunity, AI is helping to introduce a new class of molecular simulations from system size and timescale to quantum accuracy.
AI Will Create SaaS Medical Devices: The medical device industry has a game-changing opportunity enabled by AI to miniaturize and reduce cost, to automate and increase accessibility, and to continuously deliver innovation over the life of the product. This is creating a new business model to enable medical device companies to evolve from hardware solutions into software-as-a-service solutions that can be upgraded remotely and keep devices state of the art years after deployment.
AI will be Multimodal: Understanding disease is still our biggest grand challenge with over 10 thousand diseases without a therapy. Whether discovering drugs or treating patients, the use of multiple sources of health data is required. In order to leverage the world’s largest data sources with the most diversity, multimodal AI will bring us to that new frontier in discovering disease pathways as well as personalizing the treatment and prognosis of patients.
AI 2.0 with Federated Learning: To help AI application developers industrialize their AI technology and expand the application’s business benefit, AI must be trained and validated on data that resides outside the possession of their group, institution, and geography. Federated learning is the key to enable such collaboration of building robust AI models and validating models in the wild without sharing sensitive data. Federated learning will be needed at the far edges of every industry to facilitate the continuous learning and evaluation of AI.
Josh Gluck, Vice President Global Healthcare Technology Strategy at Pure Storage
A voracious appetite for faster-time-to-science is here to stay: The appetite for faster time to science is voracious and will likely continue. The world’s scientific community continues to break records in the fight against COVID-19—leveraging massive information sharing that is leading to a more accurate picture of COVID-19 and accelerated development and testing of vaccines and therapeutic treatment candidates. We’ve seen what can be done faster than ever imagined. Health sciences organizations across the board seek to build on this momentum safely and effectively to further accelerate the pace of personalized medicine. Genomics and artificial intelligence (AI) are key to this quest. To realize AI at scale, however, requires liquid data and modern data infrastructure that re-imagines the role of data and how it is used.
XNAT and self-tuning data infrastructures will transform imaging analytics to drive faster discovery and more accurate diagnostics: As healthcare and life sciences teams look to become agile with their data, the ability to automate medical imaging analytics and deploy machine learning algorithms on imaging data is critical. As a result, IT leaders at these organizations are reconsidering their data infrastructure—both storage and compute—as they rethink how to make the most of their imaging data at hand. Many of these organizations are turning to XNAT, an open-source imaging-informatics platform that helps import, archive, process, and securely distribute imaging data. Unlike PACS, XNAT has increased support for machine learning and annotation workflows. That means researchers and physicians can extend their capabilities when diagnosing diseases based on radiology. In recent years, XNAT has presented some formidable bottlenecks related to latency in data ingestion from clinical PACS to XNAT. Clinical studies on the imaging side tend to vary in their data sets, and traditional HPC storage systems are not optimized for this application. Another potential bottleneck is the learning curve in the tuning necessary to start bringing in various file types from other imaging modalities because research environments typically need to be tuned for specific data sets. The emergence of self-tuning data infrastructures accelerates data loading as well as the learning curve, unlocking the true power of XNAT to drive faster discovery and more accurate diagnostics.
Data interoperability will reach a tipping point: After years of discussion and debate, compliance deadlines for the healthcare interoperability rule and related requirements are here. The requirements, which are designed to support seamless and secure access to patient electronic health information, offer the push needed to standardize patient records and modernize legacy systems. As digitized and standardized records become available through the rule’s payer-to-payer exchange requirements, which must be met by January 2022, payers will have the ability to dig deeper into some of the social determinants of health and chronic disease that can open doors to greater patient engagement. Real-time analytics will be essential to this vision though and can only be achieved if the right foundation is in place. To realize the full potential of the interoperability rules, health care organizations need an infrastructure that is not only built for the new requirements and standards but also offers the capability to secure, process, analyze, and scale the influx of data quickly—often in real time—and in a cost-effective way.
Ashu Singhal, Co-Founder and President Benchling
75% of the global population will directly use a biotechnology product created in the past three years: Most of this will be fueled by the COVID-19 vaccine, but new fuels, foods, and agricultural products will also play a role. Biotech can now rewrite cells to cure and prevent disease, create more nutritious, higher-yield crops with built-in genetic defenses for heat and drought, and grow brand new, disruptive materials and ingredients that will reduce our dependence on petrochemicals and help solve climate change.
Look for protein sequencing to take center stage as the next breakout biotechnology: Technology advances will follow the successful path of single-cell DNA and RNA sequencing to unlock major advances in drug discovery.
Record biotech funding and exits will expand the hottest job market in America: Much like we’ve seen a talent crunch for software, look for a sizzling IPO market to drive demand and salaries across biotech.
Christopher D. Brown, CTO, 908 Devices
‘Omics’ will impact life sciences: Over the last 20 years, the explosion of the “omics” areas of knowledge discovery. It started with genomics and is now rapidly shifting to proteomics, metabolomics, and beyond. The promise of genomic information improving the understanding of healthy and aberrant biology led to the creation of an entire industry segment in tools, and that is happening again now with proteomic and metabolomic developments. These fields are going to completely change the degree of visibility we have into both state and trajectory in biological systems.
Extracting valuable insights from mountains of data: Life science researchers have incredible volumes of data coming at them daily. Many folks describe “drowning in data but starving for information.” We really need to do a much better job of extracting insights from these extremely multi-variable data sets in ways that the non-expert can quickly exploit, and at the same time ensuring that researchers have the data-oriented tools to improve the design and gathering of that data in the first place. Data science has exploded in visibility, and while there will always be a need for specialists in that field, the data gathering systems/tools/devices themselves need to put in a lot more effort to close the loop for the researcher, rather than just tossing them a lot of numbers.
Jeffrey Whitford, head of sustainability and social business innovation and branding at MilliporeSigma
Increasing developments expected in green chemistry efforts and quantifiable data collection: such as MilliporeSigma’s industry-first DOZN tool that uses the 12 principles of green chemistry for comparing the relative greenness of similar chemicals, synthetic routes, and chemical processes
Research and production applications for greener solvents and renewables will continue to increase rapidly: helping to lower a company’s carbon footprint and decreasing the risks of exposure to carcinogens and other health hazards
Increased investment in professional development for STEM educators: in hopes of improving access to science education and the quality of it to inspire the next generation of young scientists
Rapid expansion of CSR programs: for companies as they become a key driver for employee engagement
Susan Collins, global head of healthcare at Twilio
COVID-Driven AI Adoption: Healthcare burnout from COVID will be the tipping point that drives the structural change needed for widespread AI adoption by the end of 2023. We have less human capacity to deliver healthcare than we did before the pandemic, and the only way to avoid further burnout for healthcare staff is to adopt AI and tech that reduces friction and opens up clinicians’ time for meaningful care.
Varun Ganapathi, Ph.D., co-founder and CTO at AKASA
Companies will lean more on human-powered AI to avoid “Garbage In, Garbage Out” algorithms: As AI continues to evolve at a breakneck pace, companies often overlook the importance of keeping humans actively involved in the AI implementation process, creating a scenario where tech’s obsession with the newest, biggest thing neglects basics that make AI actually useful: plugging in useful data and teaching it how to deal with outliers. For AI to truly be useful and effective, a human has to be present to help push the work to the finish line. Without guidance, AI can’t be expected to succeed and achieve optimal productivity. This is a trend that will only continue to increase. Ultimately, people will have machines report to them. In this world, humans will be the managers of staff (both other humans and AIs) that will need to be taught and trained to be able to do the tasks they’re needed to do. Just like people, AI needs to constantly be learning to improve performance. A common misconception is that AI can be deployed and left unsupervised to do its work, without considering the reality that our environments are always shifting and evolving. Would a manager do this with a human worker? The answer is no.
Machine learning and human-in-the-loop approaches to automation will displace RPA: Digital transformation efforts in a number of industries have driven massive adoption of robotic process automation (RPA) during the past decade. The hard truth is that RPA is a decades-old technology that is brittle and has real limits to its capabilities—leaving a trail of broken bots which can be expensive and time-consuming to fix. RPA will always have some value in automating work that is simple, discrete, and linear. However, automation efforts often fall short of aspirations because so much of life is complex and constantly evolving—too much work falls outside of the capabilities of RPA. Emerging machine-learning-based technology platforms combined with human-in-the-loop approaches to automation are already redefining what it is possible to automate across a number of industries where complexity, exceptions, and outliers train the AI to work smarter, making automation stronger.
The AI community will go back to basics—data labeling: Solid AI systems rely on two things: a functioning model and underlying data to train that model. To build good AI, programmers need to spend the vast majority of their time collecting, categorizing, and cleaning data. For many AI technologists, a gut instinct is to run towards the sexy work of creating complex AI infrastructure and neglect the basics of data labeling.
Mark Day, EVP R&D, iRhythm
Shift from retrospective to predictive analysis: In the near future, we expect AI innovation in healthcare to shift focus from retrospective analysis to predictive insight. To reach this milestone, health wearables must become both proficient and validated in determining who needs preventive care before symptoms and associated outcome risks manifest. At its core, digital health is meant to streamline complex processes and bring preventative care to high-risk populations. Predictive AI will help to deliver on this potential.
Bias in AI: Within the next year, AI companies will continue to improve data collection methods and develop processes that avoid bias in algorithm training and, in turn, performance in the intended population. Specifically, improved clinical study design will foster more heterogeneous and representative patient populations, resulting in algorithms that reduce bias. On the technical side, methods will develop to provide greater insight into the “black box” of AI algorithm decisions, which will guide understanding into whether these decisions represent bias based on factors including race, gender, and age.
Ed Ikeguchi, CEO of AiCure
Urgency to address bias in AI: In 2022, we will need to work together across healthcare to enhance the equitability of the AI that our patients’ care increasingly relies on. AI is only as strong as the data it is fed, but today’s developers are often left to train algorithms on single-source data with limited diversity. We’ve learned that the resulting biases can impact the ability to capture accurate data around a drug’s safety, ultimately impacting patients’ health. By equipping developers with diverse, high-quality training data sets and executing more rigorous evaluations of AI’s performance, these tools can reach their full potential to truly optimize drug development and patient care.
AI-powered precision medicine: AI-powered tools will increasingly take on an important role in precisely and accurately understanding a patient’s response to treatment. Rather than relying on subjective data such as patient-reported outcomes or one-off check-ins, video and audio-based digital biomarkers can catch subtle cues of a patient’s disease progression and how a treatment impacts their daily life. The sensitivity and objectivity of these novel assessments can not only inform personalized, proactive support for a patient, but also improve a sponsor’s understanding of the right patient for their specific drug.
Reliance on open-source data platforms to advance AI: In 2022 and beyond, we will see a push to build trust in these novel measurements in the public domain through peer review. Today’s algorithms are mostly proprietary, meaning researchers cannot access them to exercise and validate them on their own data sets, limiting their potential. Open-source platforms for algorithms can allow the scientific community to collaborate and jointly contribute to novel assessments, helping them become a more widely adopted means to objectively assess patient response.
Dr. John Frownfelter, Chief Medical Officer at Jvion
Data Will Drive Action to Address Health Inequities: AI will help analyze data on social determinants of health to determine their impact on both individual patients and their communities. This will help healthcare organizations pinpoint the resources that will have the greatest impact on reducing health disparities in their communities.
AI Will Fill the Gaps Left by the “Great Resignation”: As more healthcare workers burned out by the pandemic leave the profession, AI will help health systems make the most of their increasingly scarce human resources, automating what can be automated and guiding care teams on how to best prioritize their time and resources with patients who need it the most.
Value-Based Care Will Becomes the Norm: The shift from episodic care to long-term care management will depend on preventative care and early interventions. Prescriptive analytics will help look at each individual patient holistically and identify not only the most vulnerable but also their greatest risk drivers and the interventions that will best address them.
AI Will Enable the Continued Growth of Hospital at Home: The home care model took off during the pandemic as a safer alternative to hospitals. AI will help triage patients for home care, identify potential barriers, and ingest data from monitoring devices to predict avoidable hospitalizations and recommend interventions to prevent them.