Fig. 6

Prediction comparison for STARmap. a Performance comparison via fivefold cross-validation. ENGEP (y-axis) predicts expression of measured genes (with low sparsity) and shows its superiority to benchmarked methods (x-axis) by comparing PCC values. The AR values and P values are also reported. b The predicted expression patterns of measured genes (top) and the measured patterns of them (bottom). c Correction of low-quality genes (with low sparsity). Although the predicted patterns (middle) of these genes differ from the measured ones (top), they agree well with the ISH images (bottom). d Correction of low-quality genes (with high sparsity). The predicted patterns (middle) of these high sparsity genes are mismatched with the measured patterns of them (top) but are in agreement with patterns measured by osmFISH (bottom), a spatial transcriptomics technology with high sensitivity