Research On Single Molecular Localization Algorithm For RNA 3D Imaging

As the basic unit of biological life activities,cells are the basis for human beings to understand and understand living things.Due to the large number and variety of cells in organisms,it is very difficult for human beings to understand cells.The introduction of the transcriptome provides an important scientific basis for the study of cell phenotype and function.An important part of spatial transcriptome research is to identify the types of RNA in cells and obtain the distribution of RNA in cells.Single-cell RNA sequencing,as an important technical method for transcriptome research,can sequence all RNA in tissue samples and can reflect the differences in RNA distribution between cells.However,this technology cannot achieve RNA sequencing at the subcellular scale.The emergence of single-molecule localization technology has solved the above problems.However,the bleaching of fluorescent probes brought about by this technology has become a new research direction.Biologists try to solve this problem with light sheet microscopy and three-dimensional imaging systems.However,for the RNA threedimensional imaging system,how to analyze the RNA signal and reconstruct the RNA distribution remains to be studied.Therefore,this paper mainly focuses on the three-dimensional positioning and reconstruction of RNA molecules.The main contents are as follows:(1)For the current RNA-based 3D imaging system,there is no good method to analyze the RNA signal,so that the problem of fluorescent probe bleaching caused by the single-molecule localization method cannot be solved.In this paper,we propose a method for 3D single-molecule localization based on a convolutional neural network with a network architecture consisting of 3 stacked codecs,two belonging to the feature extraction module and one belonging to the prediction module.This stacked network architecture mainly takes into account the difference in the density of fluorescent molecules in each frame of image,and the characteristics of fluorescent molecules at different densities are different.Especially for overlapping fluorescent molecules,jointly using one codec for feature extraction may make the network difficult to converge.Therefore,for fluorescent molecular images of different densities,feature extraction is performed separately according to the channel category high density and low density.The experimental results of the simulation data and the experimental data show that the three-dimensional localization method proposed in this paper has obtained a good detection rate and localization accuracy.(2)Aiming at the particularity of RNA 3D imaging data,this paper proposes a 3D reconstruction framework based on RNA localization data.According to the obtained RNA distribution in each frame of the original image,the complete RNA distribution in the sample can be obtained through the three-dimensional reconstruction framework.The framework includes three parts: coordinate transformation,3D mosaic,and deduplication of anchor points.Among them,deduplication of positioning points is a key step.In this paper,a deduplication method based on the relative distance of positioning points is proposed in combination with data characteristics such as data coordinates of positioning points.This method can quickly find out the RNA molecular points at the same position in all frames through parallel computing,and simultaneously process the RNA molecular points in adjacent frames.In order to verify the performance of the whole 3D positioning reconstruction process,experiments were carried out on the simulation data and the actual mouse brain slice data,and the reconstruction results were finally completed.The research on RNA three-dimensional signal analysis(localization and reconstruction)method in this paper is an exploration to solve the problem of fluorescent probe bleaching in single-molecule localization method,and provides a feasible solution.