scDirect: key transcription factor identification for directing cell state transitions based on single-cell multi-omics data

Chen Li, Sijie Chen, Yixin Chen, Haiyang Bian, Minsheng Hao, Lei Wei, Xuegong ZhangbioRxiv 2024 · DOI

TL;DR

This paper introduces scDirect, a novel computational framework that leverages single-cell multi-omics data (scRNA-seq and scATAC-seq) to identify key transcription factors (TFs) that drive cell state transitions. scDirect outperforms existing methods in identifying these crucial TFs and can even predict effective TF combinations for cell reprogramming, paving the way for more efficient cell engineering and regenerative medicine applications.

What Problem Does This Paper Address?

Cellular processes, from development to disease, often involve complex transitions between different cell states. While it's known that a small set of key transcription factors (TFs) can orchestrate these changes, experimentally identifying these TFs is a time-consuming and labor-intensive task. Existing computational methods often rely on bulk tissue data, which can obscure the heterogeneity of individual cells and are unsuitable for rare cell types. The advent of single-cell sequencing technologies offers a powerful new avenue, but effectively extracting TF information from this data, especially when combining different omics layers, remains a challenge. This paper aims to bridge this gap by providing a robust computational tool that can accurately identify key TFs and their combinations from single-cell multi-omics data.

Methods

scDirect employs a multi-step approach:

  1. Primary Gene Regulatory Network (GRN) Construction: It first constructs a foundational GRN using CellOracle, integrating scRNA-seq data (for regulatory coefficients) and scATAC-seq data (for regulatory directions from TFs to target genes).
  2. GRN Enhancement with Graph Attention Networks (GAT): To address missing links in the primary GRN (due to dropout events or incomplete motif databases), scDirect utilizes a GAT. This deep learning model learns to predict and recover these missing TF-target interactions, creating a more comprehensive and enhanced GRN.
  3. TF Identification as a Linear Inverse Problem: The core of scDirect models TF identification as a linear inverse problem. It uses the enhanced GRN and cell state transition vectors (derived from scRNA-seq data) to estimate the "expected alteration" of TFs that would drive the transition.
  4. Directing Score Calculation: For each TF, a "directing score" is calculated based on the Pearson correlation coefficient between its expected alteration and the overall cell state transition vector. This score quantifies a TF's importance in driving the transition.
  5. TF Combination Identification: Uniquely, scDirect can also calculate directing scores for combinations of TFs, enabling the prediction of effective reprogramming cocktails.

Datasets

The study utilizes several datasets for training and evaluation:

  • Human Embryonic Stem Cell (hESC) Atlas: A single-cell atlas generated from TF-overexpression experiments, used to establish a benchmark for TF identification. This dataset is particularly valuable as it provides experimentally validated ground truth for TF roles in differentiation.
  • Mouse Single-Cell Multi-omics Dataset: A dataset profiling mouse hair follicle development, used to demonstrate scDirect's ability to identify TFs in developmental processes.
  • Five Experimentally Validated Reprogramming Datasets: These datasets cover various cell reprogramming scenarios (e.g., fibroblasts to keratinocytes, fibroblasts to iPSCs) and were used to assess scDirect's performance in identifying TF combinations for reprogramming.

Key Insights and Ideas

  • Multi-omics Integration is Crucial: Combining scRNA-seq and scATAC-seq data significantly improves the accuracy of GRN construction and subsequent TF identification, as scATAC-seq provides crucial regulatory directionality.
  • GATs Enhance GRNs: Graph Attention Networks are effective at recovering missing TF-target links, leading to more complete and biologically meaningful GRNs. This is a key innovation that improves upon standard GRN inference methods.
  • Linear Inverse Problem Formulation: Framing TF identification as a linear inverse problem provides a principled way to infer TF activity from observed gene expression changes.
  • Directing Score as a Measure of Importance: The "directing score" offers a quantitative measure of a TF's influence on a specific cell state transition, allowing for ranking and selection.
  • Predicting TF Combinations: scDirect's ability to identify and rank combinations of TFs is a significant advancement, moving beyond identifying individual key players to predicting synergistic cocktails for complex cellular manipulations.

Key Results

  • Superior Performance: scDirect significantly outperformed existing methods (like DEA and ANANSE) in TF identification benchmarks, achieving higher TF identification scores (TIS) across multiple cell differentiation and reprogramming scenarios.
  • Effective GRN Enhancement: The GAT module improved the accuracy of TF-target link prediction, with AUROC scores ranging from 0.917 to 0.995 across different tissues.
  • Accurate Reprogramming TF Identification: scDirect successfully identified known key TFs in various experimentally validated reprogramming cases and outperformed other methods in identifying these TFs with high ranks.
  • Identification of Novel TF Combinations: The framework could quantitatively identify feasible reprogramming TF combinations, many of which were experimentally validated, suggesting its utility in accelerating experimental design. For example, in one case, the ground-truth combination ranked 25th out of nearly 10,000 combinations, and in another, it ranked 8th out of 780.

Limitations and Future Work

  • Linear Model Assumption: While effective, scDirect currently models cell state transitions as a linear process. The authors acknowledge that incorporating non-linear regulatory relationships could improve performance for more complex transitions.
  • Extension to Other Regulatory Components: The framework could be extended to include other regulatory components beyond TFs, such as chromatin regulators, splicing factors, and microRNAs, for a more holistic view of cellular control.
  • Modeling Continuous Transitions: Cell state transitions are often continuous biological processes. The authors suggest that scDirect could be extended to model TF variations as a changing tendency over time, rather than a single step.
  • Data Availability: The paper notes that while scATAC-seq data is beneficial, it's not strictly required, but its absence might impact performance. The reliability of the method is also dependent on the quality and coverage of the input single-cell data.

Summary updated

Abstract

Abstract Cell state transitions are complicated processes that occur in various life activities. Understanding and artificially manipulating them have been longstanding challenges. Substantial experiments reveal that the transitions could be directed by several key transcription factors (TFs). Here we present scDirect, a computational framework to identify key TFs based on single-cell RNA-seq and ATAC-seq data. scDirect models the TF identification task as a linear inverse problem, and solve it with gene regulatory networks enhanced by a graph attention network. Through a benchmarking on a single-cell human embryonic stem cell atlas, we demonstrate the robustness and superiority of scDirect against alternative analysis methods on TF identification. We apply scDirect on various datasets, and scDirect exhibits high capability in identifying key TFs in cell differentiation and somatic cell conversion. Furthermore, scDirect can efficiently identify TF combinations for cell reprogramming, many of which have been experimentally validated. We envision that scDirect can utilize rapidly increasing single-cell datasets to identify key TFs for directing cell state transitions and may become an effective tool to facilitate cell engineering and regenerative medicine.
BibTeX
@misc{li2024scdirect,
  author = {Chen Li and Sijie Chen and Yixin Chen and Haiyang Bian and Minsheng Hao and Lei Wei and Xuegong Zhang},
  title = {{scDirect: key transcription factor identification for directing cell state transitions based on single-cell multi-omics data}},
  year = {2024},
  doi = {10.1101/2024.01.08.574757},
  howpublished = {bioRxiv},
}
Figure