1Background: Identifying effective candidate drug compounds in patients 2 with neurological disorders based on gene expression data is of great im-3 portance to the neurology field. By identifying effective candidate drugs 4 to a given neurological disorder, neurologists would (1) reduce the time 5 searching for effective treatments; and (2) gain additional useful informa-6 tion that leads to a better treatment outcome. Although there are many 7 strategies to screen drug candidate in pre-clinical stage, it is not easy to 8 check if candidate drug compounds can be also effective to human.
9Objective: We tried to propose a strategy to screen genes whose expres-10 sion is altered in model animal experiments to be compared with gene 11 expressed differentically with drug treatment to human cell lines. 12 Methods: Recently proposed tensor decomposition (TD) based unsu-13 pervised feature extraction (FE) is applied to single cell (sc) RNA-seq 14 experiments of Alzheimer's disease model animal mouse brain. 15 Results: Four hundreds and one genes are screened as those differentially 16 expressed during Aβ accumulation as age progresses. These genes are sig-17 nificantly overlapped with those expressed differentially with the known 18 drug treatments for three independent data sets: LINCS, DrugMatrix and 19 GEO.20 Conclusion: Our strategy, application of TD based unsupervised FE, is 21 useful one to screen drug candidate compounds using scRNA-seq data set. 22 keywords: Amyloid, Alzheimer Disease, Gene Expression, Single-Cell Anal-23 ysis, Drug Discovery, Cell Line 24 1 Introduction 25 Drug discovery for neurological disorder has never been successful in spite of 26 massive efforts spent [1]. One possible reason is because we generally do not 27 1 have suitable model animals for human neurological disorder [2]. Although a 28 huge number of compounds are screened using model animals, only a few of 29 them passed the human level screening. In this sense, it is required to screen 30 candidate compounds using information retrieved from human at the earliest 31 stage. One possible strategy to do this is the usage of human cell lines; Nev-32 ertheless, it is also not easy to perform, since generating cell line from human 33 neurological disorder patients is not easy. In contrast to the cancer cell lines, 34 which can be easily generated by immortalizing tumor cells, neuronal cells are 35 hardly converted to cell lines, since mature neurons do not undergo cell divi-36 sion [3]. Therefore, it is difficult to test if candidate drugs work for human 37 during pre-clinical stages. 38 In order to overcome this difficulty, we proposed an alternative strategy; com-39 paring disease gene expression with that of compound treated animals and/or 40 human cell lines. Generally, compound screening is based upon phenotype; i.e., 41 evaluation of compounds efficiency is tested based upon if drug treatment can 42 produce symptomatic improvement. Nevertheless, since it has been recently 43 found that various neurological disorders share gene expre...