Multivariate Feature Extraction for Prediction of Future Gene Expression Profile

Introduction: The features of a cell can be extracted from its gene expression profile. If the gene expression profiles of future descendant cells are predicted, the features of the future cells are also predicted. The objective of this study was to design an artificial neural network to predict gene expression profiles of descendant cells that will be generated by division/differentiation of hematopoietic stem cells. Method: The developed neural network takes the parent hematopoietic stem cell’s gene expression profile as input and generates the gene expression profiles of its future descendant cells. A temporal attention was proposed to encode the main time series and a spatial attention was also provided to encode the secondary time series. Results: To make an acceptable prediction, the gene expression profiles of at least four initial division/differentiation steps must be known. The designed neural network surpasses the existing neural networks in terms of prediction accuracy and number of correctly predicted division/differentiation steps. The proposed scheme can predict hundreds of division/differentiation steps. The proposed scheme’ error in prediction of 1, 4, 16, 64, and 128 division/differentiation steps was 3.04, 3.76, 5.5, 7.83, and 11.06 percent, respectively. Conclusion: Based on the gene expression profile of a parent hematopoietic stem cell, the gene expression profiles of its descendants can be predicted for hundreds of division/differentiation steps and if necessary, solutions must be sought to encounter future genetic disorders.