Prediction of future gene expression profile by analyzing its past variation pattern

Area of Science:

• Biotechnology
• Computational Biology
• Genomics

Background:

• Hematopoietic Stem Cells (HSCs) possess the unique ability to self-renew and differentiate into diverse cell types.
• Understanding gene expression profiles (GEPs) during HSC division and differentiation is crucial for developmental biology and regenerative medicine.
• Predicting future GEPs of descendant cells remains a significant challenge in stem cell research.

Purpose of the Study:

• To develop a novel computational method for predicting approximate gene expression profiles (GEPs) of future descendant cells resulting from Hematopoietic Stem Cell (HSC) division and differentiation.
• To model the GEP prediction problem as a multivariate time series prediction task, incorporating correlations between GEP sequences.

Main Methods:

• Introduction of EHSCP (Extended Hematopoietic Stem Cell Prediction), an artificial neural machine designed for GEP prediction.
• Modeling GEP prediction as a multivariate time series problem, utilizing novel neural units: PLSTM (Parametric Long Short Term Memory) and MILSTM (Multi-Input LSTM).
• Implementation of a hierarchical encoder with 101 MILSTMs to compute correlations among thousands of GEP time series, enabling multi-stage prediction.

Main Results:

• EHSCP demonstrated superior prediction accuracy compared to existing methods in evaluating GEPs of descendant cells.
• The method successfully predicted 128 division/differentiation stages with high accuracy, even with only 8 initial stages provided.
• Training involved 155 datasets, with evaluation conducted on 39 independent test datasets, confirming robust performance.

Conclusions:

• EHSCP offers a significant advancement in predicting gene expression profiles for stem cell progeny.
• The developed artificial neural machine effectively captures temporal dependencies and correlations in GEP data.
• This predictive capability holds promise for applications in stem cell research, developmental biology, and therapeutic development.