With the rapid progression of biotechnologies in the last few decades, molecular biology assays have shifted from limited to high throughput. Indeed, we successively witnessed an accelerated development of 'omics technologies at multiple molecular scales: DNA arrays, DNA methylation screening, proteome-wide interaction screening, Chip-Seq, protein array, next generation sequencing, RNA-seq and next generation protein mass-spectrometry. The availability of large repository of 'omics data has stimulated the prolific growth of analytical methods for clinical outcome prediction specialized for one type of 'omics measurement, but comparatively fewer cross-scales ones (2 molecular scales) and very rare multiple scales ones (⩾3 molecular scales; "multiscale"). Of note, abundant original cross-scale and multiscale 'omics methods have been developed for identifying gene function [1], novel disease-gene [2,3] and diseases' biomodules (e.g. biomodules of microRNA-mRNA co-expression [4]). However, these approaches are insufficient for predicting clinical outcome of complex disorders [5,6]. Here, we provide a framework to illustrate the difficulty of increasing the accuracy of a clinical outcome predictor from multiple scales of 'omics data, and position the significance of Kim et al. [7] that appears in this issue of the Journal of Biomedical Informatics. We also include a brief historical perspective of foundational methodologies in cross-scale and multiscale 'omics analytics (single 'omics analytics are out of scope of this perspective).