Identification Of Mouse Circadian Rhythmic Phenotype With Convolutional Neural Networks

Circadian rhythm refers to a physiological phenomenon in which an organism changes its activity in a cycle of about 24 hours.When the biological rhythm is destroyed,many physiological processes and behaviors are abnormal.Therefore,studying the biological circadian rhythm is of great significance.In the paper,we used large-scale data on the activity and diet of mutant mice,using convolutional neural networks to identify circadian rhythm phenotypes of mouse activity and diet.The main work of the thesis includes the following two parts:(1)Convolutional neural network and training set construction;data in the database established by IMPC is identified by convolutional neural network,but since the data morphology of most mice in the database is consistent with the data morphology of wild-type mice,Missing phenotypic data,most of the data is biased towards normal.Therefore,the data is biased and it is not possible to directly select part of the data in the IMPC database as a training set.Next,we used the characteristics of the real experimental data of the diet and activity of the mice to artificially construct the training set.Then,using two layers of convolutional neural networks for training,a predictive convolutional neural network model can be obtained.(2)Mouse phenotype recognition;the diet and activity of mice usually assume the shape of two peaks,the first peak of which is the main peak and the height is higher.We use two peak height anomalies as indicators of abnormal biological circadian rhythms.A convolutional neural network was used to automatically identify two peak height inverted phenotypes.Several genes were found in the diets and activities of the four centers,and the phenotype of mice lacking these genes showed a characteristic of peak inversion.The innovation of the thesis is to provide a method for artificially generating a feasible training set for the large deviation of the data set and the lack of the labeled training set,and construct the corresponding convolutional neural network for training.In the paper,convolutional neural networks are used to automatically discriminate phenotypic genes with abnormal rhythm.This method provides the possibility to identify large-scale rhythmic data,and has made qualitative progress with non-automated judgment technology.This paper designs a synthetic training set and uses convolutional neural network to automatically,efficiently and accurately identify a parameter in large-scale biological data,and also provides a design idea for the machine learning algorithm without missing labeled training set.The genes screened in this paper also provide candidate targets for further biological experiments.