[论文解读] Sequence models for continuous cell cycle stage prediction from brightfield images

L. Léger, Maxine Leonardi+3 more|ArXiv.org|Feb 4, 2025

Cell Image Analysis Techniques被引用 3

一句话总结

该论文评估因果状态空间与 transformer 序列模型，以从无标签的明场图像预测连续的 Fucci 信号，效果优于单帧基线，并实现如 G1/S 等一小时分辨率的转变。

ABSTRACT

Understanding cell cycle dynamics is crucial for studying biological processes such as growth, development and disease progression. While fluorescent protein reporters like the Fucci system allow live monitoring of cell cycle phases, they require genetic engineering and occupy additional fluorescence channels, limiting broader applicability in complex experiments. In this study, we conduct a comprehensive evaluation of deep learning methods for predicting continuous Fucci signals using non-fluorescence brightfield imaging, a widely available label-free modality. To that end, we generated a large dataset of 1.3 M images of dividing RPE1 cells with full cell cycle trajectories to quantitatively compare the predictive performance of distinct model categories including single time-frame models, causal state space models and bidirectional transformer models. We show that both causal and transformer-based models significantly outperform single- and fixed frame approaches, enabling the prediction of visually imperceptible transitions like G1/S within 1h resolution. Our findings underscore the importance of sequence models for accurate predictions of cell cycle dynamics and highlight their potential for label-free imaging.

研究动机与目标

• 通过明场成像推动无标记的细胞周期分析，以推断连续的细胞周期状态。
• 提供一个可公开获取的大规模、带有地面实况 Fucci 信号的跟踪细胞数据集。
• 系统比较单帧、因果与非因果序列模型在细胞周期预测中的性能。
• 量化在正常和干扰条件下的预测准确性与像G1/S等转变的时序精度。

提出的方法

• 使用 ResNet-18 主干从明场图像中提取每帧嵌入。
• 将嵌入输入到共享约100万参数的不同序列模型头。
• 比较单帧、固定历史的因果 CNN、状态空间（Mamba）与基于 Transformer 的架构。
• 使用 L1 损失和动态时间规整（DTW）来衡量信号准确性与时序对齐。
• 在完整轨迹（M-M）、部分轨迹以及药物干扰数据（Palbociclib）上进行评估。
• 对 Transformer 使用轮换位点嵌入，并在训练时采用常见的数据增强。

Figure 1: Multi-modal imaging of Fucci -reporter cells reveals a continuous representation of cell cycle states. a) Time-lapse imaging of Fucci-reporting cells allows for precise quantification of cell cycle staging through the characteristic oscillations of fluorescent reporter intensities. b) Repr

实验结果

研究问题

• RQ1 是否能在没有荧光报告剂的情况下，通过时间序列的明场图像预测连续的 Fucci 信号？
• RQ2 时间模型（因果与非因果）在追踪细胞周期转变时是否优于单帧方法？
• RQ3 模型选择如何影响关键转变（如 G1/S、S/G2）以及药物干扰下的预测？
• RQ4 在轨迹不完整或细胞被药理干扰时，预测是否具备鲁棒性？

主要发现

• 序列模型在从明场图像预测 Fucci 信号方面优于单帧基线。
• 双向 Transformer 在 DTW 基于的对齐中表现最佳，L1 误差最低。
• 状态空间模型（Mamba）显著优于固定历史的因果 CNN，表明更长范围的时序传播有益。
• 基于明场的预测接近 H2B 荧光输入的性能，凸显明场数据中关于细胞周期的强形态信息。
• Transformer 能准确捕捉药物治疗细胞的扰动，双向模型在分布外情形下最具鲁棒性。
• 部分轨迹中，因果模型保持较好准确性，尽管对非常短的片段优势减弱。

Figure 2: Overview of prediction approach. a) We use a ResNet-18 [ He et al.(2016)He, Zhang, Ren, and Sun ] to extract single frame embeddings from a brightfield sequence which are fed into a sequence model that predicts both Fucci channels. b) Sequence models explored in this paper: Single Frame ML