Deep Learning-Based Prediction of Time-Series Single-Cell RNA-Seq Data

Document Type

Conference Article

Publication Title

Lecture Notes in Networks and Systems

Abstract

Single-cell transcriptomics or RNA-sequencing (scRNA-seq) technology enables unprecedented molecular profiling and provides novel insights into cellular heterogeneity and cell states, which bulk RNA-sequencing fails to capture. Transcriptomics data profiled at multiple time-points during developmental or disease stages, also known as time-series scRNA-seq, is now being increasingly available. Analysis of such data allows extraction of informative genes as compared to a static snapshot drawn at a single time-point. However, most of the present-day models ignore the time dimension within the data. This limits accurate inferencing of cellular trajectory and states encoded by the underlying genes. Utilizing the available time component could yield useful information regarding the underlying mechanisms at play during development and disease. Forecasting gene expressions for an advanced time-point is also necessary when data is degradable or missing. Hence, in this work, we have attempted to develop a deep neural network (DNN)-based prediction model for estimating gene expression values in time-series scRNA-seq data. The DNN regressor is capable of estimating data at advanced time-points of interest once trained on multiple previous time-points. The performance of the proposed model compares favorably with some state-of-the-art regression models that exist at present.

First Page

213

Last Page

226

DOI

10.1007/978-981-19-6791-7_13

Publication Date

1-1-2023

This document is currently not available here.

Share

COinS