Fig. 1
From: A pitfall for machine learning methods aiming to predict across cell types
The performance of neural network models of varying complexity in three predictive settings on two tasks. Schematic diagrams of a cross-chromosome, b cross-cell type, and c hybrid cross-cell type/cross-chromosomal model evaluation schemes. d–f The figure plots the average precision (AP) of a machine learning model predicting gene expression as a function of model complexity. Evaluation is performed via d cross-chromosome, e cross-cell type, and f a combination of cross-chromosome and cross-cell type validation. In each panel, each point represents the test set performance of a single trained model. g–i is the same as d–f but predicting TAD boundaries rather than gene expression