Sequence models for continuous cell cycle stage prediction from brightfield images

TL;DR

This work demonstrates that continuous cell-cycle states can be inferred from label-free brightfield images by leveraging temporal sequence models. By building a large, annotated RPE1 Fucci dataset and comparing single-frame, causal state-space, and transformer architectures, the authors show that sequence models significantly outperform static baselines and can capture subtle transitions like G1/S with high temporal resolution. The bidirectional transformer achieves the best overall performance, while causal state-space models offer strong, potentially real-time capabilities; the study also reveals robustness limitations under drug perturbations and emphasizes the need for diverse data and uncertainty estimation for broad applicability.

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.

Figures (12)

• 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) Representative time-lapse images of brightfield, H2B, and Fucci channels across one full (M-M) cell cycle. c) Quantification of integrated logarithmic fluorescence intensities from Fucci reporters, normalized nuclear area, and normalized total H2B signal in a representative full M-M (mitosis to mitosis) track. Vertical lines mark the time points corresponding to the images shown in d. d) Log-transformed Fucci manifold for continuous inference of cell cycle states.
• Figure 2: Overview of approach.a) We use a ResNet-18 he2016deep to extract single frame embeddings from an input sequences which are fed into a sequence model that predicts both Fucci channels. b) Sequence models explored in this paper: Single Frame MLP, Fixed-frame CNN, causal state-space models e.g. Mamba gu_mamba_2024, bidirectional models e.g. transformers vaswani_attention_nodate.
• Figure 3: Predictions on unperturbed RPE cells Regular.a) Distribution of $L1$ errors across the different models. b) Predictions of Fucci signals on two example tracks: one with accurate and one with poor predictions. The ground truth signal is shown in black. c) Average prediction error, and d) variability of ground truth Fucci signals as a function of normalized cell cycle time $\tau$.
• Figure 4: Comparative performance on partial cell cycle tracks (brightfield). Shown is the average $L_1$ error of both Fucci signals when using partial tracks as input, parametrized by their relative start and end time $\tau_1 \leq \tau_2 \in [0,1]$.
• Figure 5: Results on perturbed RPE cells Drug.a,b) Effect of CDK4/6 inhibition on cell cycle durations and onset of G1, S, and G2/M phases, extending G1 duration while leaving S and G2/M unchanged. c) Example Fucci predictions on Drug.