TY - JOUR TI - Network-Based Prediction of Drug-Combination Efficiency Based on Expression Data on Off Target Genes AU - Ligeti, Balázs AU - Menyhárt, Otilia AU - Juhász, János AU - Pongor, Sándor AU - Győrffy, Balázs T2 - Jedlik Labor Rep AB - Gene expression data from public databases can be efficiently incorporated into algorithms predicting drug combination efficiency by using a generalized framework of network neighborhoods. In this representation, efficient drugs are those where the affected targets overlap well with the known disease genes, and synergistic drugs are those whose neighborhoods overlap well with the purported disease genes. Drug combinations are very efficient and frequently used tools in the treatment of complex multigene diseases such as hypertension, diabetes and cancer. Broadly speaking, the effect of combining pharmacons can be either synergistic (beneficial), or additive (no interactions) or antagonistic (negative, adverse) and delineating the effect of combination therapies must include careful scrutiny of these possibilities. Nevertheless, most currently used combinations were found in tedious empirical ways, consequently there is a growing interest for finding potentially interacting drug partners using in silico methods. One of the potential avenues is the network propagation hypotheses which posits that the genes or proteins, included those affected by pharmacons, are interlinked by an interaction network, which is perturbed by the action of drugs. It is supposed that perturbations can propagate (diffuse) via the links of this network and combination effects will result in unexpected accumulation of perturbations at various points of the network. The gross effect of drug combination as the portion of jointly perturbed genes which gives a simple [0.1] measure, the so called Target Overlap Score (TOS), TOS was shown to correlate well both with known drug/combination data and with the results of newly conducted clinical trials. However there are also limitations, such as the lack of distinction between positive and negative correlations, and the difficulty of introducing experimental data into the system. In this work we plan to integrate experimental data available in public databases into network based prediction of drug interactions. The idea is to use a generalized framework of network interactions in which both pharmacon target genes and disease-affected genes are symbolically represented as off-target gene neighborhoods. Known off-target genes can also be added as separate neighborhoods. In the previous studies, the neighborhoods were computationally generated by calculating the diffusion of perturbation around a known target or off-target gene. Here we make use of experimentally determined neighborhoods extracted from a database of gene expression data, mainly microarray experiments. It is also possible to combine the two principles (ie. using calculated and experimentally derived neighborhoods) by calculating the diffusion neighborhood around experimentally validated of target genes. This generalization may allow one to incorporate Validation of the new principle can be carried out by using public data taken from the DCDB, TTD and Drugs.com databases, or using published clinical data. Preliminary studies show that incorporating gene expression data improves the correlation, so we trust that this principle can further improve the efficiency of predictive algorithms. DA - 2019/// PY - 2019 VL - 7 IS - 1 SP - 75 EP - 76 J2 - Jedlik Laboratories Reports SN - 2064-3942 UR - https://m2.mtmt.hu/api/publication/30762840 ER -