2016 Fiscal Year Annual Research Report
| Project/Area Number |
16J02715
|
|---|
| Research Institution | Hokkaido University |
|---|
Principal Investigator |
TESSMER HEIDI LYNN 北海道大学, 獣医学研究科, 特別研究員(DC2)
|
|---|
| Project Period (FY) |
2016-04-22 – 2018-03-31
|
| Keywords | infectious disease / machine learning / mathematical modeling / epidemiology |
|---|
| Outline of Annual Research Achievements |
This year's research was primarily focused on learning and exploring new machine learning techniques and exploring their application to epidemiological and genetic data.
To control epidemics from infectious diseases the estimation of the basic reproduction number, R0, and is very important. This study explores the ability of a machine to estimate R0 and other epidemiological model parameter values of a disease given a time series of observed incidences obtained from a mathematical model describing the transmission of infectious disease, known as the SEIR model. Traditionally, Approximate Bayesian Computation (ABC) techniques have been used to estimate these parameters, however these techniques have several limitations. We explore a machine learning approach, using a simple Multi-Layer Perceptron (MLP) to learn and estimate the parameter values which addresses some of the limitations of the ABC technique. Preliminary experiments suggest that the machine learning approach is useful for estimating R0. We expand this approach to include the other parameters of SEIR models and, further, compare the accuracy and time requirements for both techniques on both test and real-world datasets.
Collaborating with other students, a new method to track the evolution of influenza A viruses during interspecies transmission between avian and swine hosts and a new method to estimate the lineage dynamics of human influenza B viruses were developed.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
Through collaborative research with the Seoul National University Biointelligence Laboratory, I greatly expanded my knowledge and understanding of machine learning as a whole, culminating in acceptance of a paper to the Machine Learning for Health Workshop at the Neural Information Processing Systems (NIPS) Conference in December. The title of my accepted submission was ‘Estimation of Basic Reproduction Number R0 using a Recurrent Neural Network.’
The year's work also included co-authorship on two papers relating to the mutation of influenza viruses, including the human influenza B virus dual-lineage prediction analysis (Nyirenda, et al 2016) and avian and swine interspecies transmission of influenza A viruses (Karnbunchob, et al 2016).
|
| Strategy for Future Research Activity |
First, I will complete the work exploring machine learning estimation of epidemiological model parameters and its comparison to the traditional approximate Bayesian computation method. Next, I will focus on completing my thesis describing this and previous work on machine learning and research into interspecies transmission of viruses. Finally, I would like to expand the research back into analysis of host and viral DNA and explore classification techniques via machine learning.
|
