This study demonstrates that deep learning, combined with nonlinear dimension reduction, can reconstruct biological processes from raw image data. The researchers applied this approach to reconstruct the cell cycle of Jurkat cells and disease progression in diabetic retinopathy. They showed that deep learning can detect and separate a subpopulation of dead cells in an unsupervised manner and classify cell cycle stages with a sixfold reduction in error rate compared to a previous method. Deep learning-based predictions are fast enough for on-the-fly analysis in an imaging flow cytometer. A major challenge in biology is interpreting high-throughput single-cell data. The study focuses on imaging data from fluorescence microscopy, particularly from imaging flow cytometry (IFC), which combines the fluorescence sensitivity and high-throughput capabilities of flow cytometry with single-cell imaging. IFC is well-suited for deep learning due to its high sample numbers and multi-channel image data. Deep learning can process the increased information content in IFC data compared to conventional flow cytometry. Deep learning improves data analysis for high-throughput microscopy compared to traditional machine learning. This is mainly due to three advantages: no need for preprocessing, improved prediction accuracy, and visualizable features. The study shows that deep learning can reconstruct continuous biological processes, which has been a focus of research. Only one recent work on deep learning in high-throughput microscopy discusses the visualization of network features, but none deal with continuous biological processes. The study shows that deep learning can reconstruct the cell cycle progression of Jurkat cells. The neural network's last layer activations were used to visualize and organize single-cell data. The data was organized in a long stretched cylinder along which cell cycle phases are ordered. The DNA content of cells was used to color the data, showing the continuous progression of cells in G1, S, and G2. The study also detected abnormal cells in an unsupervised manner, showing that the neural network can learn clusters of abnormal cells independently of cell-cycle-label based training. Deep learning outperforms boosting for cell classification. The study showed that deep learning achieves high accuracy and precision, leading to an almost diagonal confusion matrix. Deep learning can classify all seven cell cycle stages with high accuracy. The study also reconstructed disease progression in diabetic retinopathy, showing that the network can order disease states along severity without being provided with the ordering information. The study shows that deep learning can reconstruct continuous biological processes based on categorical labels. This is possible because adjacent classes are morphologically more similar than classes that are temporally further separated. The study also shows that deep learning can detect abnormal cells, which has practical applications in cell analysis. The study highlights the advantages of deep learning in cell classification, including its speed and ability to process large cell populations. The study concludes that deep learning can be helpful for understanding a wide variety of biological processes involving continuous morphology changes.This study demonstrates that deep learning, combined with nonlinear dimension reduction, can reconstruct biological processes from raw image data. The researchers applied this approach to reconstruct the cell cycle of Jurkat cells and disease progression in diabetic retinopathy. They showed that deep learning can detect and separate a subpopulation of dead cells in an unsupervised manner and classify cell cycle stages with a sixfold reduction in error rate compared to a previous method. Deep learning-based predictions are fast enough for on-the-fly analysis in an imaging flow cytometer. A major challenge in biology is interpreting high-throughput single-cell data. The study focuses on imaging data from fluorescence microscopy, particularly from imaging flow cytometry (IFC), which combines the fluorescence sensitivity and high-throughput capabilities of flow cytometry with single-cell imaging. IFC is well-suited for deep learning due to its high sample numbers and multi-channel image data. Deep learning can process the increased information content in IFC data compared to conventional flow cytometry. Deep learning improves data analysis for high-throughput microscopy compared to traditional machine learning. This is mainly due to three advantages: no need for preprocessing, improved prediction accuracy, and visualizable features. The study shows that deep learning can reconstruct continuous biological processes, which has been a focus of research. Only one recent work on deep learning in high-throughput microscopy discusses the visualization of network features, but none deal with continuous biological processes. The study shows that deep learning can reconstruct the cell cycle progression of Jurkat cells. The neural network's last layer activations were used to visualize and organize single-cell data. The data was organized in a long stretched cylinder along which cell cycle phases are ordered. The DNA content of cells was used to color the data, showing the continuous progression of cells in G1, S, and G2. The study also detected abnormal cells in an unsupervised manner, showing that the neural network can learn clusters of abnormal cells independently of cell-cycle-label based training. Deep learning outperforms boosting for cell classification. The study showed that deep learning achieves high accuracy and precision, leading to an almost diagonal confusion matrix. Deep learning can classify all seven cell cycle stages with high accuracy. The study also reconstructed disease progression in diabetic retinopathy, showing that the network can order disease states along severity without being provided with the ordering information. The study shows that deep learning can reconstruct continuous biological processes based on categorical labels. This is possible because adjacent classes are morphologically more similar than classes that are temporally further separated. The study also shows that deep learning can detect abnormal cells, which has practical applications in cell analysis. The study highlights the advantages of deep learning in cell classification, including its speed and ability to process large cell populations. The study concludes that deep learning can be helpful for understanding a wide variety of biological processes involving continuous morphology changes.