The role of music in brain-computer interfaces for emotion mediation…
In February 2020, PLOS published a Collection entitled “Mathematical Modeling of Infectious Disease Dynamics” which includes papers from PLOS ONE, PLOS Biology and PLOS Computational Biology, on a variety of topics relevant to the modeling of infectious diseases, such as disease spread, vaccination strategies and parameter estimation. As the world grappled with the effects of COVID-19 this year, the importance of accurate infectious disease modeling has become apparent. We therefore invited a few authors featured in the Collection to give their perspectives on their research during this global pandemic. We caught up with Verrah Otiende (independent researcher, Pan African University Institute of Basic Sciences Technology and Innovation), Lauren White (USAID), Jess Liebig (CSIRO) and Johnny Whitman (The Ohio State University) to hear their reflections on this collection and the time that has passed.
In this second blog post of two, we hear from Jess Liebig and Johnny Whitman, who discuss the modeling of human movement, the assumptions that go into creating a model, the virtue of simpler models, and the importance of understanding under-reporting in disease modeling.
What is your research focused on currently?
JL: Since September 2017 I am part of CSIRO’s DiMeMo (Disease Networks and Mobility) team. The aim of DiNeMo is to understand how human infectious diseases might arrive and spread in Australia. We analyse various sources of data and identify patterns of people movement both internationally and domestically in order to forecast the risk of disease spread. Initially I worked on modelling dengue importations via air travel. However, since the beginning of the pandemic the focus of my work has shifted to COVID-19. I am currently studying the effects of international travel restrictions on COVID-19 importation risk. The results of this study shed light onto how many importations a country can expect when opening its borders and can guide authorities in making decisions.
JW: My research is currently split between two main thrusts: the first is a collaboration with Battelle Memorial Institute, working on comparisons of codon usage in certain classes of proteins. The second is investigating methods of identifying parameters in biological signaling networks using the supercomputing cluster at Nationwide Children’s Hospital in Columbus, Ohio. Finally, I am finishing my PhD this spring semester and my current research for that deals with the design and verification of biological circuits for intracellular signaling, as well as developing methods to coarse-grain out complicated host-virus interactions in simulations of dendritic and epithelial cells.
What do you think are the lessons we can learn from the research in your field which will help us to better model infectious diseases in the future?
JL: We need high quality datasets to accurately model the spread of infectious diseases. In reality, the datasets that are accessible for researchers are often biased, incomplete and erroneous. While the process of data collection can be tedious and expensive it can add much value to the research community when done in an organised and purposeful manner.
JW: A trend in current modeling is to hyperfocus on fitting parameters in a model in order to precisely match available data; with advances in artificial intelligence and neural networks, researchers are quick to use these overparameterization models to get very good fits to the data. I would argue that we should instead focus on identifying important qualitative features of data or populations – a difficult and careful human process – and implementing simpler models around these features. To be concrete, if a complex model of American COVID-19 cases from January to May fits the data extremely well, but offers 500 parameters to change to predict future behavior, it is very difficult to make any form of meaningful prediction or understanding of what the model is actually saying about the underlying population, whereas a simpler model with directly interpretable parameters may perform worse quantitatively, but be much more expressive overall.
Have your motivations, direction or the way you conduct or disseminate your research changed in 2020 as a consequence of the COVID-19 pandemic, either for yourself or the field as a whole?
JL: My work is motivated by several studies that have shown that the structure of the global air transport network as well as the increasing volume of international travellers has contributed to the large-scale spread of infectious diseases. The COVID-19 pandemic is an unpalatable reminder of human movement being able to rapidly spread a disease across the globe. While the motivation and direction of my work has been reinforced as a consequence of the pandemic, there have been changes to the way I disseminate my research. With travel restrictions and lockdowns in place, conferences, research meetings etc. have moved online, giving rise to new challenges. For example, it can be more difficult to clearly communicate your ideas to collaborators in a teleconference as opposed to a face-to-face meeting. What I find particularly challenging is to give online presentations where you cannot see the reaction of your audience.
JW: I think the pandemic has (or should have) focused researchers more on making observations in real populations and taking note of how real behavior patterns can make fundamental difference in model predictions. A simple example is a very good group at the University of Illinois put together an intricate and well-thought out model, which ultimately failed. The failure was due to not including the possibility that a contagious individual who knew they were contagious would continue to be social. Clearly, they are not at fault for using a rational actor assumption, but the lesson is that we should always remain grounded in the people and phenomenon we model if we hope to make any progress.
If there was one thing you wished that the general public understood better about modeling infectious diseases, what would that be?
JL: Naturally, when modelling the spread of infectious disease (or any other process), scientists have to make certain assumptions due to incomplete data and knowledge gaps. It is very important to understand what exactly these assumptions are and how they affect the results of the modelling study. Any conclusions have to be drawn carefully, taking into consideration the set of assumptions that were made.
JW: Partially due to the manner in which models are presented to the public and also how researchers have positioned their work, I think that the public believes that models are intended to exactly predict the course of a disease. Rather, I wish we collectively understood the role of modeling more as a probe into the possibilities of a system; I would never trust a model to truly predict the number of COVID cases, but they can give us the possibilities of recurrent infection waves, how the dynamics depend on observable parameters like recovery time and incubation period, and other broad qualitative features that can influence public health decisions. A more technical wish would be that the public understood model predictions in the same sense that they understand weather predictions; most complex systems modeling is stochastic in some sense, so I would prefer that reporting on modeling emphasized the possibilities of events more than definitive statements. We’ve seen public support unnecessarily erode due to unrealized model predictions, and I think this could be avoided if communication was clearer.
Are there any unanswered research questions in this field that you would really like to see us make progress on?
JL: A key ingredient to modelling the spread of infectious disease is the incidence rate. Unfortunately, the incidence of most infectious diseases is under-estimated, which is due to under-reporting and under-ascertainment. Under-reporting refers to positive disease cases not being reported, for example due to mis-diagnosis. Under-ascertainment occurs when infected individuals do not report to a health professional, for example due to the absence of symptoms. Reporting and ascertainment rates vary across time and space and depend on the disease itself. A model that requires incidence rates as input can only be accurate if we have a good understanding of the level of under-estimation surrounding the incidence rates. Unfortunately, current techniques for determining the level of under-estimation are time consuming, expensive and often biased.
JW: The physics background in me would like to see a more general study of disease modeling in the spirit of field theory models; due to the much simpler nature of interactions in theoretical physics problems, we have done a careful and systematic investigation of how essentially every class of interaction type affects the macroscopic behavior of the model, e.g. if there is some symmetry, what types of particles are allowed, if this interaction is strong, it suppresses that behavior. I would like to see a similar-minded effort in disease modeling, so that researchers in this community build up a common base of tools and understanding. As it stands, the field is so fragmented in terminology and approach that it is difficult to quickly agree about what the setup of a problem is, much less the implications of the model.
About the authors:
Jess Liebig: Jessica Liebig is a postdoctoral fellow at the Commonwealth Scientific and Industrial Research Organisation (CSIRO), Australia’s national science agency. She received a BSc(Hons) and a PhD in Applied Mathematics from RMIT University in 2013 and 2017, respectively. Her primary research interest lies in the area of network science and is directed towards the study of infectious disease spread. She is part of CSIRO’s Disease Networks and Mobility (DiNeMo) project, an interdisciplinary research initiative that aims to understand how human infectious diseases might arrive and spread in Australia. As part of her work she identifies patterns of people movement, both internationally and domestically, to forecast the risk of disease spread.
Johnny Whitman: John Whitman graduated from the University of Illinois in 2016, and is currently finishing his PhD in Physics at The Ohio State University with Prof. Ciriyam Jayaprakash. His research interests include stochastic modeling of systems at all scales, from intracellular signaling pathways to large scale population epidemiological modeling. He is most interested in problems which exhibit some form of complexity, since he really enjoys scientific programming and visualization/animation of processes.
Disclaimer: Views expressed by contributors are solely those of individual contributors, and not necessarily those of PLOS.
Featured Image : Spencer J. Fox, CC0