Figure 1 | Scientific Reports

Figure 1

From: Interpretable unsupervised learning enables accurate clustering with high-throughput imaging flow cytometry

Figure 1

Deep unsupervised learning-based single-cell clustering workflow. (i) After the sample preparation, cells are examined using the 3D-IFC system. (ii) The deep unsupervised learning model takes cell images as inputs and is trained to encode the inputs into latent representations. t-SNE visualization of the trained results of latent space representations. (iii) Model-predicted labels were generated at the clustering step. t-SNE visualization of the clustering result. (iv) For post-evaluation, the model-generated labels are compared with the ground truth labels, and the confusion matrix is produced. qRT-PCR can also be used for downstream analysis after the clustering step. AOD, acousto-optic deflector; CL, cylindrical lens; IO, 20 × /0.42 illumination objective; SDO, 10 × /0.28 side detection objective; SSP, side spatial filter; DMs, dichroic mirrors; FDO, forward detection objective; FSP, forward spatial filter; PMT, photomultiplier tube.

Back to article page