Candidate Mechanism Perturbation Amplitude Algorithm

NPA have been developed by the team of Prof. Manuel Peitsch (Philipp Morris International, PMI). NPA algorithms work directly on BEL mechanistic model graphs; they can directly be applied to identify patient subgroups. In our case we applied the Candidate Mechanism Perturbation Amplitude (CMPA) algorithm, which explores the combined effect of numerous differentially expressed genes by quantifying the magnitude of impact on biological process: a down-regulation would yield a low CMPA score and alternatively, up-regulation would result in a high CMPA score. The algorithm is based on the concept of heat diffusion. The process mimics the diffusion of heat from “leafs” of the network to neighbors through the entire network: In our case “heat” is the gene expression values of the genes (“leafs”). The obtained scores are compared across different conditions such as brain regions or patient age groups or disease stages to identify conditions that best represent a biological process, or in this case condition setting, a disease mechanism. The scores can be evaluated by findings in literature, or tested through wet-lab experiments (e.g., by applying drug interventions to revert the observed effects). In this manuscript, we explain how CMPA algorithm can be used in NeuroMMSig subgraphs to elucidate novel insights in the context of the mechanisms studied in AETIONOMY.

The CMPA algorithm explores the combined effect of numerous differentially expressed genes by quantifying the magnitude of impact on biological process: a down-regulation would yield a low CMPA score and alternatively, up-regulation would result in a high CMPA score. The algorithm is based on the concept of heat diffusion. The process mimics the diffusion of heat from “leafs” of the network to neighbours through the entire network: In our case “heat” is the gene expression values of the genes (“leafs”). The obtained scores are compared across different conditions such as brain regions or patient age groups or disease stages to identify conditions that best represent a biological process, or in this case condition setting, a disease mechanism. The scores can be evaluated by findings in literature, or tested through wet-lab experiments (e.g., by applying drug interventions to revert the observed effects). In this deliverable, we present how CMPA algorithm can be used in NeuroMMSig subgraphs to elucidate novel insights in the context of the mechanisms studied in AETIONOMY.

The CMPA algorithm necessarily requires a causal network i.e. relationships between nodes should be either ‘increases’ or ‘decreases’. This can be achieved by filtering and removing non-causal relationships in the NeuroMMsig sub-graphs. The next step is a) performing differential gene expression analyses b) extract relevant gene expression subsets to map them on the nodes of the network. Next step is: Calculation of the biological impact factor (BIF) on nodes that have incoming edges using the formula described above (see figure above). The aggregated BIF scores represent the behaviour of a regulated mechanism. i.e. Higher the BIF score is higher is the perturbation.

The CMPA algorithm was implemented for 3 different mechanisms. i.e.:

  • Mitochondrial dysfunction in PD
  • NFT aggregation in AD and
  • Neuroinflammation in AD.

The magnitude of mitochondrial dysfunction in PD was calculated for datasets GSE57475 (PD age-group data), GSE49036 (Braak Staged patients) and GSE28894 (brain specific expression data).

Similarly, datasets GSE28146 and GSE1297 are both disease stages specific and were used for NFT aggregation and Neuroinflammation in AD respectively.

Figure: A brief outline of the CMPA algorithm