Cell reprogramming design by transfer learning of functional transcriptional networks

Area of Science:

• Synthetic Biology and Computational Biology
• Machine Learning in Genomics
• Cellular Network Dynamics

Background:

• Designing cell reprogramming treatments faces challenges due to incomplete cellular network knowledge and vast intervention possibilities.
• Existing experimental methods struggle with the complexity of predicting cellular responses to gene perturbations and drug treatments.

Purpose of the Study:

• To develop a computational approach for rationally designing cell reprogramming strategies.
• To overcome limitations in understanding cellular networks and combinatorial interventions for therapeutic development.

Main Methods:

• A transfer learning approach was developed, pre-trained on human cell fate transcriptomic data to model network dynamics.
• The method integrates transcriptional responses to gene perturbations to guide cells between specified initial and target states.
• Validated on large-scale microarray and RNASeq datasets covering numerous cell types and perturbations.

Main Results:

• The approach achieved high accuracy (AUROC of 0.91) in reproducing known cell reprogramming protocols.
• Demonstrated an adaptable pre-trained model capable of tailoring to specific cell fate transitions.
• Quantified the relationship between developmental relatedness and the number of gene perturbations required for cell fate steering.

Conclusions:

• The developed transfer learning method provides a proof-of-concept for computationally designing effective cell control strategies.
• Offers valuable insights into gene regulatory network governance of cellular phenotypes.
• Paves the way for rational design of novel disease treatments through cell reprogramming.