Fig. 3: Understanding molecular basis of protein prediction.
From: Inferring protein expression changes from mRNA in Alzheimer’s dementia using deep neural networks
a Ablation experiment to identify key data types. The clei2block model was trained with the removal of entire input variables of each data type as indicated in the heatmap. The average Pearson’s correlations between predicted and actual protein levels for each model are indicated in the bar graph on the right. b Influential predictors for protein abundance. The predictive contribution of each variable in the model with mRNA-PCs and all LM inputs was estimated using GradientExplainer. Representative gene ontologies enriched for genes contributing to the influential PCs were described for each PC. The enrichment analysis was conducted using one-sided Fisher’s exact test. The significance levels were adjusted for multiple comparisons using FDR at 5%. c Molecular characteristics defining protein predictability. Correlations of protein with mRNA and estimated protein and the difference between these were compared with characteristics in the protein measurement and molecular interactions.
