FatePredictor: Cell fate decision-making prediction with an ensemble deep learning model

Public summary

* FatePredictor can predict cell fate bifurcations and classify bifurcation types from locally observed data.

* FatePredictor employs dynamic unbalanced optimal transport method to reconstruct dynamic cell trajectories.

* FatePredictor achieves higher accuracy in identifying critical transitions compared with existing methods.

* FatePredictor pinpoints key genes and pathways involved in essential biological processes.

Abstract

During cell differentiation, there typically exists a drastic and sudden shift called cell fate decision-making or bifurcation. Revealing such critical phenomena can provide deeper insights into the fundamental mechanisms that govern the complex intricacies of living organisms. However, many conventional statistical methods fail to predict the specific types of critical transitions and accurately infer cell fate dynamics from single-cell RNA sequencing data. To address this challenge, we develop FatePredictor, a novel computational framework grounded in bifurcation theory and optimal transport theory, to predict cell fate bifurcation based on locally observed information of single-cell data. Specifically, the proposed FatePredictor employs a dynamic unbalanced optimal transport method to reconstruct dynamic cell trajectories, based on which an ensemble deep learning model is utilized to predict the type of dynamics involved in a cell fate bifurcation during cellular processes. The applications on both simulated and real single-cell data demonstrate that FatePredictor serves as a user-friendly and powerful tool for predicting bifurcations of complex biological systems and unveiling intricate cellular trajectories, with higher accuracy compared with many existing methods. Additionally, our FatePredictor has the capacity to pinpoint key genes and pathways related to significant cellular processes.