An AI-Based Evolution of Clinical Communities
December 13, 2019 | Enkelejda Miho believes in artificial intelligence. The technology can bring efficacious, precise, and effective diagnostics and therapeutics faster to the patient, she says, and in the future, more efficient operations could save money, freeing up capital to invest in innovation.
Miho is a professor for Digital Life Sciences at School of Life Sciences, FHNW University of Applied Sciences and Arts Northwestern Switzerland. She holds degrees in Pharmaceutical Chemistry and Technology from the University of Bologna (2011) and Pharmaceutical Medicine from the European Center of Pharmaceutical Medicine (ECPM), University of Basel (2016). She obtained her doctorate degree in Biotechnology from ETH Zurich (2017).
As a Pioneer Fellow at the Innovation & Entrepreneurship Lab (ETH ieLab), she founded aiNET, the immuno-informatics ETH spin-off for therapeutic antibody and T-cell discovery.
On behalf of Bio-IT World, Kent Simmons, a Senior Conference Director at Cambridge Healthtech Institute, spoke with Miho about the benefits of AI, the data volume challenge, and educating the next generation of data scientists. Their conversation has been edited for length and clarity.
Bio-IT World: What are the benefits that biopharmaceutical research organizations will gain by implementing machine learning and AI as they transition from traditional computational modeling?
Enkelejda Miho: Biopharmaceutical companies benefit from artificial intelligence and machine learning along the entire pipeline of drug discovery and development. While previous informatic and other statistical methods were focused on modeling typically low-dimensional systems, artificial intelligence (AI) can discover patterns in complex data. Machine learning advances take advantage of the large-scale data produced in high-throughput technologies like screening and sequencing. These methods support target selection, hit discovery, and lead optimization by reducing material cost and labor. For example, deep learning has been used for regulatory genomics where variations between individuals are considered. Neural networks exploit training datasets from intra-individual variations to predict the effect of mutations in silico and uncover regulatory effects of rare SNVs. This is an important step also for patient stratification and precision medicine.
Another application involves prediction of chemical toxicity, or drug-target prioritization. On the drug development and diagnostic side, applications are focused in diagnosis, patient stratification, and monitoring. An example is the role of artificial intelligence in identifying diabetic retinopathy from retinal images of diabetic patients with the same or greater sensitivity than that of ophthalmologists. With digital transformation and the implementation of electronic health records (important for retrospective real-world clinical data) and digital biomarkers, artificial intelligence has the potential to drastically improve healthcare, for example in neuro.
Therefore, artificial intelligence can bring efficacious, precise and effective diagnostics and therapeutics faster to the patient. It makes operations more efficient and less costly, and investments can be re-directed on non-automatable tasks, innovation and actual creation.
All R&D organizations will have issues to confront related to increasing volumes of data, the standardization of data from different sources and the protocols of capturing data in the right way going forward. This seems like an enormous challenge.
This is a global challenge that needs the coordination of several stakeholders. Industry has a crucial role as the standardization of data volumes as the economic weight of the establishment and especially long-term maintenance of such complex initiatives is not feasible for governmental agencies and public funds. Latest efforts are investigating the application the intrinsic power of AI to detect patterns in non-standardized data. These developments might bring a paradigm shift in the direction of efforts at the global level. However, potential and validation of these powerful tools are at present far apart.
How will these companies deal with implementing AI and machine learning from a systems standpoint? Will these tools require new investments in software, hardware and staff?
The implementation of AI needs specific expertise, similar to when a company is establishing any new technology in house. In addition, challenges on the diversity of the tools available, the limited transparence on the interpretation of outputs and algorithms, ethical questions, and the need for a very close contact between different specialty experts (e.g., chemists/biologists and data scientists/software engineers) adds an additional layer of complexity. Companies are trying to team up with complementary companies and this strategy can give a false feeling of security of having assured access to the innovation potential of AI. However, knowledge and communication gaps will need to be bridged. I believe these technologies need to be implemented in teams, not cross-disciplinary as traditional education imposes, but adisciplinary where motivated experts are further trained and developed in new technologies.
At your institution, the University of Applied Sciences and Arts Northwestern Switzerland, you have just introduced a graduate level program in medical informatics. How will this program support the staffing needs of pharma companies as they implement AI and machine learning?
The FHNW University of Applied Sciences and Arts in Northwestern Switzerland is an applied university, traditionally very close to industry and implementation of products. As such, the need expressed from concentrated biopharma established especially in Basel and in Switzerland was clearly perceived and action was taken. The School of Life Sciences and School of Business have established and coordinate a Master of Science program in Medical Informatics. FHNW is a first mover in addressing an educational program that combines education in digital transformation healthcare, economics, and most importantly, hands-on programming and informatics. Students are put in contact early on projects and interested biopharmaceutical companies. On the other side, additional lecturers are selected because of their industry-expertise in topics that are relevant to biopharma today. The setting works for both parties early on and leverages project specificity (while theoretical generalities are kept succinct).
How do you imagine the relationship of the research and clinical communities evolving as digital medicine becomes more widespread?
The relation of research and clinical communities is intensifying by the day. Digital medicine has transformed the very definition of health. The traditional definition of “health vs. disease” (but what is disease?) has changed to health vs. the average population, and with the advent of precision and personalized medicine (N-of-1 trials) to the health of the individual vs. self, the patient becoming the e-patient (“e”: educated, engaged, enabled, and empowered). Ultimately, the patient becoming the citizen, where his/her health is managed continuously, starting from prevention.
Clinical communities have the responsibility of generating high-quality data and carefully regarding standards during these mobile and dynamic times, where information is sensible and corruptible. This transformation requires significant research efforts in terms of the interpretability of the new findings, and investigation of analytical methods suited for novel tasks. Therefore, research communities become essential to investigating hypotheses that are generated at the speed of data, interpreting new findings, educating and communicating not only to the scientific and biopharma community but especially to clinicians and citizens.