Abstract:
The number of single-cell datasets has been increasing rapidly thanks to the technologies developed in recent years. Specialized cell types, biological cause-effect relat...Show MoreMetadata
Abstract:
The number of single-cell datasets has been increasing rapidly thanks to the technologies developed in recent years. Specialized cell types, biological cause-effect relationships, and disease mechanisms can be revealed with the use of these datasets containing omic measurements at cellular resolution together. Single-cell datasets of the same type are used together by horizontal integration. For this process, first of all, the datasets are aligned by associating the similar cells they contain. Aligned datasets are transfered to a common domain by removing the batch effect noises they contain. The accuracy of the corrected measurements is highly dependent on the accuracy of the cell matches that make up the alignment. Horizontal integration algorithms commonly use the MNN algorithm for cell matching. This method, which is based on mutually similar local neighborhood relationships, does not take into account the neighborhood relationship across the datasets, causing bias. In order to overcome this problem, a two-stage hierarchical matching algorithm is proposed in this study. In the developed method, each data set is clustered within itself and similar cell subsets in different data sets are matched with each other. Thus, similar cells are brought together by using both the general neighborhood information of each dataset and the common neighborhood information contained across the datasets. Cell matches are obtained by running the MNN algorithm independently on the connected components formed by the paired cell clusters. The performance increases significantly when the general neighborhood information is included to the matching. The results obtained in the comparative experiments show that the proposed method can match cells of the same type with 2% higher accuracy than its widely used state-of-art competitors. Moreover, it is observed that the developed method correctly matches 20 times more cells than the standard MNN algorithm. The obtained results show that the ...
Date of Conference: 05-08 July 2023
Date Added to IEEE Xplore: 28 August 2023
ISBN Information:
Print on Demand(PoD) ISSN: 2165-0608