A systems approach to discover new therapeutics
Rapidly decreasing costs of molecular measurement technologies not only enable profiling of disease sample molecular features (e.g., transcriptome, proteome, metabolome) at different levels (e.g., tissues, single cells), but also enable measuring of molecular signatures of individual drugs in clinically relevant models. Exploring methods to relate diseases to potentially efficacious drugs through various molecular features is critically important in the discovery of new therapeutics. The target-based drug discovery approach that focuses on interfering with individual targets is challenged by the lack of drug efficacy, drug resistance, and off-target effects. We propose to employ a systems-based approach that identifies drugs that reverse the molecular state of a disease. Using this approach, we have successfully identified drug candidates for three cancers: Ewing’s sarcoma (Pessetto, Oncotarget, 2016), liver cancer (Chen, Gastroenterology, 2017) and basal cell carcinoma (Mirza, JCI Insight, 2017) in the past few years. In the Ewing sarcoma work, this systems-approach achieved a hit rate of >50% in predicting effective drugs. In the liver cancer work, we demonstrated that the expression of disease genes was reversed in a clinically relevant mouse model after drug treatment. In our recent pan-cancer analysis, we found that the potency to reverse cancer gene expression correlates to drug efficacy (Chen, Nature Communications, 2017). Moving forward, we will move our drug candidates to the clinic for the liver cancer patients, meanwhile, we will explore the following topics.
1. Identify drugs to overcome drug resistance. Our past focus was on primary cancers, which have a large amount of molecular data publicly available. Since drug resistance is emerging in various cancers and relevant data are growing rapidly, we are excited to apply the systems approach to study drug resistance. For example, in neoadjuvant chemotherapy (NAC), breast cancer patients who achieve pathologic complete response (pCR) likely have long-term survival, yet there remains a significant number of patients with poor response to NAC and who fail to achieve pCR. New drugs are therefore needed to overcome chemoresistance to help patients achieve pCR. We propose to identify drugs that reverse the signature of non-responders.
2. Discover novel oncogene inhibitors. Despite the discovery of numerous driver oncogenes, many of the most prevalent oncogenic alterations cannot be blocked with small molecule inhibitors. We will use the systems-based approach to identify small molecules that reverse oncogene specific signatures. We are working on c-MYC and several others.
3. Develop the Open Cancer Therapeutic Discovery portal (OCTAD) To help computational biologists contribute to the therapeutic discovery pipeline improvement, we have released our desktop version. To help researchers easily use our pipeline to predict drug candidates for the disease of their interest, we are actively developing a web portal.
Personalized cancer therapy
Current preclinical and clinical approaches select therapies primarily based on actionable mutations, yet patients may have no actionable mutations or multiple actionable mutations that are hard to prioritize, suggesting the need for additional biomarkers. The recent disappointing result from the SHIVA trial indicates that more predictive biomarkers of drug efficacy are needed to select therapies. In addition to mutations, other molecular data (e.g., gene, protein) have been widely explored as biomarkers in personalized therapy, including gene expression based biomarkers such as Oncotype DX and MammaPrint in breast cancer treatment, however, the biomarkers for many cancer drugs are currently quite limited as it takes many years to run clinical trials to identify and validate a biomarker for a single drug. The recent efforts have enabled the large-scale identification of various types of molecular biomarkers through correlating drug sensitivity with molecular profiles of pre-treatment basal cancer cell lines. We are interested in developing methods to match these biomarkers to individual patients to inform therapy in the clinic. Our first effort is centered on building statistical models to improve drug sensitivity data (Zhiyue Hu, 2019, PSB).
A big data approach to study metastatic cancer
Metastasis is the most common cause of cancer-related death, as such, there is an urgent need to discover new therapies to treat metastasized cancers. We are extending our primary cancer work to study metastatic cancer. We are interested to use open big data to ask significant questions related to metastatic cancer research. One recent example is to evaluate cell lines and organoids as models for metastatic cancer.
Deep learning and Deep reinforcement learning in drug discovery
The new technologies in deep learning and deep reinforcement learning are simply blowing our mind everyday. We are interested in implementing these technologies into drug discovery. Examples include using deep learning autoencoder to learn gene expression profiles (William Zeng, BMC Genomics, 2018) and exploring GAN to design new compounds. Too much to do! If you understand this figure, please join us:)
Drug mechanisms in a cellular context
One challenge in the systems-based therapeutic discovery approach is that drug mechanisms are barely known. We will leverage drug-target prediction models we previously developed ( 1, 2, 3) and the new drug/shRNA-mediated gene expression profiles to infer mechanisms of actions in individual cell lines.