Motivation: Medulloblastoma (MB) is the most common malignant brain tumor in children.Despite aggressive therapy, about one-third of patients with MB still die, and survivors suffer severe long-term side effects due to the treatments. The poor post-treatment outcomes are tightly linked to unpredictable drug resistance. Therefore, before developing robust single drug or drug combination recommendation algorithms, uncovering the underlying protein-protein interaction (PPI) network patterns that accurately explain and predict drug resistances for MB subtypes is essential and important.
Results:In this study, we hypothesize that the loop sub-structure within the PPI network can explain and predict drug resistance. Both static and dynamic models are built to evaluate this hypothesis for three MB subtypes. Specifically, a static model is created to first validate that many reported therapeutic targets are located topologically on highly deregulated loop sub-structure and then to characterize the loop for tumors without treatment. Next, with the after-treatment timeseries genomics data, a dynamic hidden Markov model (HMM) with newly designed initialization scheme estimates the successful and unsuccessful occurrence probabilities for each given PPI and then re-delineates the loop for post-treatment tumors. Finally, the comparison of loop structures pre-and post-treatment distinguishes effective and ineffective treatment options, demonstrating that the loop sub-structure is capable of interpreting the mechanism of drug resistance. In summary, effective treatments show much stronger inhibition of cell cycle and DNA replication proteins when compared to ineffective treatments after considering the cross talk of multiple pathways (the loop).