Heatmap from Scientific Research

Using fewer but more predictive modalities generally led to improved performance for most models. Model performance in terms of mean Antolinis C and mean IBS across splits as a function of the number of modalities used for training: models fitted on clinical data and gene expression compared to all modalities. The dark color indicates that the model in question performed better for a particular dataset when trained on all modalities compared to only clinical data and gene expression modalities used as input. (C) Comparison for Antolinis concordance.

DOI