Predicting outcome of Morris water maze test in vascular dementia mouse model with deep learning

PLOS ONE, Nov 2019

The Morris water maze test (MWM) is one of the most popular and established behavioral tests to evaluate rodents’ spatial learning ability. The conventional training period is around 5 days, but there is no clear evidence or guidelines about the appropriate duration. In many cases, the final outcome of the MWM seems predicable from previous data and their trend. So, we assumed that if we can predict the final result with high accuracy, the experimental period could be shortened and the burden on testers reduced. An artificial neural network (ANN) is a useful modeling method for datasets that enables us to obtain an accurate mathematical model. Therefore, we constructed an ANN system to estimate the final outcome in MWM from the previously obtained 4 days of data in both normal mice and vascular dementia model mice. Ten-week-old male C57B1/6 mice (wild type, WT) were subjected to bilateral common carotid artery stenosis (WT-BCAS) or sham-operation (WT-sham). At 6 weeks after surgery, we evaluated their cognitive function with MWM. Mean escape latency was significantly longer in WT-BCAS than in WT-sham. All data were collected and used as training data and test data for the ANN system. We defined a multiple layer perceptron (MLP) as a prediction model using an open source framework for deep learning, Chainer. After a certain number of updates, we compared the predicted values and actual measured values with test data. A significant correlation coefficient was derived form the updated ANN model in both WT-sham and WT-BCAS. Next, we analyzed the predictive capability of human testers with the same datasets. There was no significant difference in the prediction accuracy between human testers and ANN models in both WT-sham and WT-BCAS. In conclusion, deep learning method with ANN could predict the final outcome in MWM from 4 days of data with high predictive accuracy in a vascular dementia model.

A PDF file should load here. If you do not see its contents the file may be temporarily unavailable at the journal website or you do not have a PDF plug-in installed and enabled in your browser.

Alternatively, you can download the file locally and open with any standalone PDF reader:

https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0191708&type=printable

Predicting outcome of Morris water maze test in vascular dementia mouse model with deep learning

February Predicting outcome of Morris water maze test in vascular dementia mouse model with deep learning Akinori Higaki 0 1 Masaki Mogi 1 Jun Iwanami 0 1 Li-Juan Min 0 1 Hui-Yu Bai 0 1 Bao- Shuai Shan 0 1 Masayoshi Kukida 0 1 Harumi Kan-no 0 1 Shuntaro Ikeda 1 Jitsuo Higaki 1 Masatsugu Horiuchi 0 1 0 Department of Molecular Cardiovascular Biology and Pharmacology, Ehime University, Graduate School of Medicine , Tohon, Ehime , Japan , 2 Department of Cardiology, Pulmonology, Hypertension and Nephrology, Ehime University, Graduate School of Medicine , Tohon, Ehime , Japan , 3 Department of Pharmacology, Ehime University, Graduate School of Medicine , Tohon, Ehime , Japan 1 Editor: Manabu Sakakibara, Tokai University , JAPAN The Morris water maze test (MWM) is one of the most popular and established behavioral tests to evaluate rodents' spatial learning ability. The conventional training period is around 5 days, but there is no clear evidence or guidelines about the appropriate duration. In many cases, the final outcome of the MWM seems predicable from previous data and their trend. So, we assumed that if we can predict the final result with high accuracy, the experimental period could be shortened and the burden on testers reduced. An artificial neural network (ANN) is a useful modeling method for datasets that enables us to obtain an accurate mathematical model. Therefore, we constructed an ANN system to estimate the final outcome in MWM from the previously obtained 4 days of data in both normal mice and vascular dementia model mice. Ten-week-old male C57B1/6 mice (wild type, WT) were subjected to bilateral common carotid artery stenosis (WT-BCAS) or sham-operation (WT-sham). At 6 weeks after surgery, we evaluated their cognitive function with MWM. Mean escape latency was significantly longer in WT-BCAS than in WT-sham. All data were collected and used as training data and test data for the ANN system. We defined a multiple layer perceptron (MLP) as a prediction model using an open source framework for deep learning, Chainer. After a certain number of updates, we compared the predicted values and actual measured values with test data. A significant correlation coefficient was derived form the updated ANN model in both WT-sham and WT-BCAS. Next, we analyzed the predictive capability of human testers with the same datasets. There was no significant difference in the prediction accuracy between human testers and ANN models in both WT-sham and WT-BCAS. In conclusion, deep learning method with ANN could predict the final outcome in MWM from 4 days of data with high predictive accuracy in a vascular dementia model. - Funding: The authors received no specific funding for this work. Competing interests: The authors have declared that no competing interests exist. Introduction Dementia is becoming a serious health problem throughout the world, along with the aging of society. Since the importance of dementia study increases, the use of experimental methods to evaluate cognitive function is also increasing. The Morris water maze test is one of the most popular and established behavioral tests to evaluate rodents' spatial learning and memory, which was originally invented by Richard G. Morris in 1983 [ 1 ]. This test has been widely used not only in the area of neuroscience but also in the field of cardiovascular study as the concept of the neurovascular unit became prevalent. However, its application for mice has some disadvantages, because this test was calculated for rats. First of all, the overall performance is lower in mice than in rats. In addition, their performance is susceptible to their swimming ability, vision and motivation. As a result, mice show floating and thigmotaxis more frequently than rats [ 2, 3 ]. This test also has some issues for testers. Although this method is a useful way to evaluate cognitive function, it requires a substantially longer experimental period. Considering the protocols used in recent studies, the standard training period in this test is about five days [4]. However, this period is not a fixed value and is left to the researcher's discretion. In fact, the results of the final experimental day often seem self-evident to experienced testers. Therefore, we assumed that if the result on the final day can be estimated with high accuracy, it may be possible to shorten the experimental period and reduce physical and mental burden on testers. Deep learning is a class of machine learning techniques which uses multiple layers of nonlinear information processing to recognize feature quantities in data [ 5 ]. An artificial neural network (ANN) is one of the architecture to realize deep learning and has been applied for a variety of tasks including medical and pharmaceutical research [ 6, 7 ]. The major advantage of ANN is that we can easily obtain a highly accurate mathematical model from a certain amount of dataset without knowing the detailed internal process [8]. There (...truncated)


This is a preview of a remote PDF: https://journals.plos.org/plosone/article/file?id=10.1371/journal.pone.0191708&type=printable

Akinori Higaki, Masaki Mogi, Jun Iwanami, Li-Juan Min, Hui-Yu Bai, Bao-Shuai Shan, Masayoshi Kukida, Harumi Kan-no, Shuntaro Ikeda, Jitsuo Higaki, Masatsugu Horiuchi. Predicting outcome of Morris water maze test in vascular dementia mouse model with deep learning, PLOS ONE, 2018, Volume 13, Issue 2, DOI: 10.1371/journal.pone.0191708