Transcriptomic forecasting with neural ordinary differential equations.
Transcriptomic forecasting with neural ordinary differential equations.
复制标题
DOI：
10.1016/j.patter.2023.100793
复制
发表时间：
2023-08-11
期刊：
影响因子：
6.5
通讯作者：
Fertig, Elana J.
中科院分区：
文献类型：
--
作者：
Erbe, Rossin;Stein-O'Brien, Genevieve;Fertig, Elana J.
单细胞转录组学技术可以揭示细胞表型背后的分子状态变化。然而，理解动态细胞过程需要从细胞状态的快照推断轨迹扩展到估计细胞基因表达的时间变化。为了应对这一挑战，我们开发了一种基于神经常微分方程的方法RNAForecaster，用于以嵌入独立的方式预测单个细胞中未来多个时间步的基因表达状态。我们证明，RNAForecaster可以准确地预测未来的表达状态，在模拟的单细胞转录组数据与细胞跟踪随着时间的推移。然后，我们表明，通过使用来自组成性分裂细胞的代谢标记单细胞RNA测序（scRNA-seq）数据，RNAForecaster准确地重现了3天内细胞周期进展过程中基因表达的许多预期变化。因此，RNAForecaster能够从具有时间信息的高通量数据集对生物系统中的未来表达状态进行短期估计。神经ODE方法预测单细胞的未来基因表达状态我们证明了在训练数据中未观察到的表达状态的预测性能通过跟踪3天内的细胞周期基因表达来验证，对于基因表达动态的估计不需要减少重复性单细胞转录组学数据产生细胞表达水平的快照，此时细胞在测序之前被杀死。每个细胞的时间背景的缺乏阻碍了分析，因为在大多数条件下，细胞的基因表达水平预计是恒定的。已经做出了一些努力来从单细胞RNA测序数据推断时间信息，例如伪时间和RNA速度方法。这些技术依赖于将数据投影到低维空间中，并且仅在由此观察到的表达空间内进行预测。我们开发了RNAForecaster模型，它预测计数空间中的表达水平，并试图推广到所有可能的表达水平，而不仅仅是训练数据中观察到的表达水平。我们证明了RNAForecaster在模拟和实验数据集中生成准确的短期到中期RNA表达预测。目前用于从单细胞转录组学数据推断时间信息的方法依赖于诸如UMAP的降维方法，并且只能在训练数据中捕获的低维空间内进行预测。我们提出了一种基于神经ODE的方法来预测未来的表达状态，并证明了在不使用降维或限制基于训练数据的表达水平预测的情况下，单细胞表达水平的准确的短期到中期预测是可能的。
Single-cell transcriptomics technologies can uncover changes in the molecular states that underlie cellular phenotypes. However, understanding the dynamic cellular processes requires extending from inferring trajectories from snapshots of cellular states to estimating temporal changes in cellular gene expression. To address this challenge, we have developed a neural ordinary differential-equation-based method, RNAForecaster, for predicting gene expression states in single cells for multiple future time steps in an embedding-independent manner. We demonstrate that RNAForecaster can accurately predict future expression states in simulated single-cell transcriptomic data with cellular tracking over time. We then show that by using metabolic labeling single-cell RNA sequencing (scRNA-seq) data from constitutively dividing cells, RNAForecaster accurately recapitulates many of the expected changes in gene expression during progression through the cell cycle over a 3-day period. Thus, RNAForecaster enables short-term estimation of future expression states in biological systems from high-throughput datasets with temporal information. Neural ODE method predicts future gene expression states of single cells We demonstrate prediction of expression states not observed in training data Performance is validated by tracking cell cycle gene expression over 3 days Dimensionality reduction is not required for estimation of gene expression dynamics Single-cell transcriptomics data yield snapshots of cell expression levels at the moment the cells were killed prior to sequencing. The lack of temporal context for each cell hampers analysis because, under most conditions, the gene expression levels of a cell are expected to be in constant flux. Several efforts have been made to infer temporal information from single-cell RNA sequencing data, such as pseudotime and RNA velocity methods. These techniques rely on projecting the data into a lower-dimensional space and only predicting within the expression space thus observed. We have developed the RNAForecaster model, which predicts expression levels in the counts space and attempts to generalize to all possible expression levels rather than just those observed in the training data. We demonstrate that RNAForecaster generates accurate short- to medium-term RNA expression predictions in simulated and experimental datasets. Methods currently used to infer temporal information from single-cell transcriptomics data rely on dimensionality reduction methods such as UMAP and can only predict within the lower-dimensional space captured in the training data. We propose a neural ODE-based method for predicting future expression states and demonstrate that accurate short- to medium-term predictions of single-cell expression level are possible without using dimensionality reduction or restricting expression level predictions based on the training data.
登录
查看更多内容
影响因子：
56.9
作者：
Battich, Nico;Beumer, Joep;van Oudenaarden, Alexander
通讯作者：
van Oudenaarden, Alexander
影响因子：
64.5
作者：
Dixit, Atray;Pamas, Oren;Li, Biyu;Chen, Jenny;Fulco, Charles P.;Jerby-Amon, Livnat;Marjanovic, Nemanja D.;Dionne, Danielle;Burks, Tyler;Raychowdhury, Raktima;Adamson, Britt;Norman, Thomas M.;Lander, Eric S.;Weissman, Jonathan S.;Friedman, Nir;Regev, Aviv
通讯作者：
Regev, Aviv
影响因子：
64.8
作者：
La Manno G;Soldatov R;Zeisel A;Braun E;Hochgerner H;Petukhov V;Lidschreiber K;Kastriti ME;Lönnerberg P;Furlan A;Fan J;Borm LE;Liu Z;van Bruggen D;Guo J;He X;Barker R;Sundström E;Castelo-Branco G;Cramer P;Adameyko I;Linnarsson S;Kharchenko PV
通讯作者：
Kharchenko PV
影响因子：
64.5
作者：
Qiu, Xiaojie;Zhang, Yan;Martin-Rufino, Jorge D.;Weng, Chen;Hosseinzadeh, Shayan;Yang, Dian;Pogson, Angela N.;Hein, Marco Y.;Min, Kyung Hoi (Joseph);Wang, Li;Grody, Emanuelle, I;Shurtleff, Matthew J.;Yuan, Ruoshi;Xu, Song;Ma, Yian;Replogle, Joseph M.;Lander, Eric S.;Darmanis, Spyros;Bahar, Ivet;Sankaran, Vijay G.;Xing, Jianhua;Weissman, Jonathan S.
通讯作者：
Weissman, Jonathan S.
影响因子：
48
作者：
Qiu Q;Hu P;Qiu X;Govek KW;Cámara PG;Wu H
通讯作者：
Wu H