Fig. 1: Machine learning techniques used in MnM. | Nature Communications

Fig. 1: Machine learning techniques used in MnM.

From: Unravelling single-cell DNA replication timing dynamics using machine learning reveals heterogeneity in cancer progression

Fig. 1

a, b Copy number imputation with k-Nearest Neighbors (KNN). Single-cell copy number data in a matrix or a BED file are used as input and the missing copy number values marked as a question mark (a) are filled in by KNN imputation (b). c Deep learning model for the single-cell replication state classifier. The trained deep learning model includes one input layer and three hidden layers. The output layer is loaded and used to predict the replication states of single cells. d–f Subpopulation discovery in three steps. Dimensionality reduction is performed with UMAP and non-replicating cells in two dimensions to provide representative lower dimensions of the copy number data (d). DBSCAN clusters the data based on the UMAP coordinates (e). This allows matching replicating cells to the corresponding non-replicating subpopulations with KNN after a second 10-dimension UMAP dimensionality reduction step (f).

Back to article page