Drug Combination Prediction Based On Complex Networks

The drug combination can cover multiple drug targets,effectively perturbing the network,enhancing drug efficacy,and reducing side effects,which is important in complex diseases such as tumors and pathogenic infections.However,there are many drawbacks to drug combination discovery methods: firstly,some methods ignore drug indications or consider only a single disease;secondly,most methods explore 2-drug combinations and are not applicable to multi-drug combinations;and finally,some data-driven methods are difficult to interpret their rationale and results.Firstly,a network model PINet 1.0 is constructed based on the full range of human genes and pathways.Both diseases and drugs are extrasystematic perturbation factors for the network,which makes the model logically applicable to multiple diseases.Second,drugs are mapped to their corresponding targets,and multi-drug combinations are simulated by linear summation of targets.Third,The effects of drugs and diseases on the network are compared,and when they can counter each other,the drug combination is suitable for the disease and may make synergistic effects,which can improve the model interpretation to some extent.Finally,the drug combination space is reduced according to the mechanism of the drug combination when synergy occurs,significantly reducing computational time.Firstly,the sensitivity of PINet 1.0 to different diseases was assessed and the model was suitable for acute myeloid leukemia,inflammatory bowel disease,breast cancer and non-small cell lung cancer,but not for diabetes,AIDS and hyperlipidemia.Secondly,the sensitivity of PINet 1.0to different drug combinations was assessed and the model was suitable for drug combinations from 2 to 5.Finally,drug combinations for acute myeloid leukemia were predicted by PINet 1.0.Among the predicted results,gemtuzumab ozogamicin and midostaurin being indicated for newly diagnosed acute myeloid leukemia,which has passed clinical phase I trials and is proposed for phase II trials.PINet 1.0 is a high-order drug combination prediction model with some explanatory power for multiple diseases,which is independent of the training set.The model is constructed using the modularity principle,and the sensitivity of the model to infectious diseases can be subsequently improved by introducing pathogen modules.The mechanism of drug combinations can also be explained by refining the network modulation.In addition,this study discusses the potential application of PINet 1.0 in the field of disease classification and traditional Chinese medicine formulation optimization.