Finding combinatorial drug repositioning therapy for Alzheimer's disease and related dementias
Alzheimer's disease (AD) is increasing in number as the U.S. and global population gets older. Despite enormous effort and expense. treatment remains at best modestly effective and no new drugs have been approved since 2003 [1.2]. The increasing interest in drug discovery through repositioning old drugs for new therapeutic applications throw light in the treatment for neurodegenerative diseases including AD [3]. Drug repositioning has unique advantages in speeding up the development process with a much lower cost and safer pathway (repositioned drugs have less risk of unexpected toxicity or side effects) when compared to de novo drug discovery approaches [4]. However. this is a non-trivial task and naive in silico methods for drug design and discovery fails to identify useful targets [5]. AD is a highly complex disease characterized by distinct and neuropathological phenotypes [6]. and there are even syndromes mimics AD (amnestic dementia syndrome [7]). which was recently identified in retrospective autopsy studies. Because our understanding on AD is still not comprehensive. many seemingly promising results (based on partially validated mechanisms) show no activity in vivo. leading to their abandonment in further studies [3]. We believe it is necessary to think out of the box and increase the search space to include drug combinations. which can overcome drug resistance and target multiple biomarkers for more effective treatment. Depending on the perspective. we can exam this problem either bottom-up or top-down. The bottom-up strategy is to screen drug combinations based on their causal effects to AD using observed healthcare data (claims. electronic health records. etc.). The top-down strategy. on the other hand. is based on the pharmacology and biology knowledge (i.e.. drug-gene interaction and cellular signaling pathways). There is unique strength in each strategy and we will carefully study them in the following aims: Aim 1: Identify potential drug combinations that could lead to a lower Alzheimer onset rate using large observational databases (e.g.. EHR and claim database) Task 1.1: Develop highly scalable algorithms based on balanced gradient boosting tree to mine drug combinations that have positive causal relationship with the low onset rate of AD Task 1.2: Develop novel optimization strategies based on tensor factorization to identify clusters of candidate drugs to reduce the search space for drug combinations while maintaining high model utility Task 1.3: Use cell-model to test and validate the efficacy of highly ranked drug combination candidates Aim 2: Predict AD-protective drug combinations based on cellular signaling pathways Task 2.1: Develop gene oncology knowledgebase based on literature to predict drug combinations targeting specific genes Task 2.2: Use oncology enriched multi-modality gene disease network (including drug. gene. pathway. and disease) to predict potential AD-protective drug combinations using deep graph convolution network (GCN) Task 2.3: Conduct biological validation on combinatorial drug therapies in ???????in vitro???????? models of AD Aim 3: Develop software and webservice to enable user-friendly interactions and efficient deployment Task 3.1: Build ready-to-use client-side toolkits. and open-source them through Github and release preconfigured Docker container in Dockerhub to enable easy adoption Task 3.2: Develop a webservice with RESTful APIs to support cloud-side applications by moving computation to data rather than supplying data to model We will combine big biomedical data from complementary sources. state-of-the-art informatics models. AD-specific domain expertise. and biology knowledge and validation into a coherent framework to tackle AD with potential combinatorial drug therapy in an exponentially larger. challenging yet promising space when compared with traditional drug repositioning models considering one single drug at a time.