The direction of links in a network is important

The networks shown on this site describe the most likely regulation around individual genes in relationship to Alzheimer’s disease (AD) phenotypes. These networks are unusual in that directions are assigned to each link in the network. These directed links place molecules as either “upstream” of an AD phenotype (molecule → phenotype), or “downstream” of a phenotype (phenotype → molecule). The implication of a molecule that is upstream of a phenotype (molecule → phenotype) is that when that gene’s expression changes, then the AD phenotype will change in response. Perturbing the expression of downstream genes is predicted to NOT have an effect on the AD phenotype. Similarly, arrows between biological regulators describe their causal relationships in the vicinity of a specific gene.

Distinguishing between upstream and downstream molecules is important for identifying chains of activation that are commonly involved in biological processes. For instance, if you observe 1000 differentially expressed genes in persons with AD, which of those are driving the disease progression, and which of those downstream consequences of the disease? Correctly determining upstream vs downstream molecules in these situations may assist in identifying influential genes that drive expression of many other genes and overall disease processes, as opposed to spending time on molecules related to the downstream consequences of the disease. The genes downstream of pathology may also be important and related to the symptoms of a disease, but are NOT predicted to control the origin of disease or molecular cascades related to the disease. This strict definition of upstream and downstream molecules is necessarily false in many cases, due the presence of loops in molecular networks. However, classifying molecules as upstream or downstream of pathology is practically useful in drug development, where upstream targets are considered desirable.

The networks presented here are those which best fit all of the multi-omic and phenotype data recorded for several hundred people in the ROSMAP aging cohorts, who are in various stages of health and disease. Therefore, the accuracy of these networks is limited by the accuracy and scope of the input data. Also, we do not or cannot measure all forms of biological regulation—for instance 3D chromatin state is not yet measured—so there are certain to be inaccuracies in at least some models. However, the ROSMAP cohort has extensive molecular and clinical phenotypes, so these results are reasonably comprehensive. Whether a gene of interest is labeled as upstream or downstream, please note that we simply presenting the probabilistic results that best fit these particular data and not categorically claiming that this is the biological truth, but rather a useful categorization in conjunction with other measures in prioritizing AD-relevant genes.