Technological breakthrough in Deep Learning (DL), part of Artificial Intelligent (AI), has revolutionized industries, automation and also biomedical research. While its impact starts to be felt in our daily life, whether they are AI enabled consumer products or autonomous driving cars, many of these sophisticated deep learning algorithms have not yet been adapted to harness knowledges from ever-increasing genomic data, such as those from Cancer Genome Atlas (TCGA, >10,000 tumors across 33 tumor types), Cancer Cell-Line Encyclopedia (CCLE, >1000 cell lines), and enormous therapeutic drug testing results. Recently, we have developed a novel machine learning model to predict drug response...

Read More

Technological breakthrough in Deep Learning (DL), part of Artificial Intelligent (AI), has revolutionized industries, automation and also biomedical research. While its impact starts to be felt in our daily life, whether they are AI enabled consumer products or autonomous driving cars, many of these sophisticated deep learning algorithms have not yet been adapted to harness knowledges from ever-increasing genomic data, such as those from Cancer Genome Atlas (TCGA, >10,000 tumors across 33 tumor types), Cancer Cell-Line Encyclopedia (CCLE, >1000 cell lines), and enormous therapeutic drug testing results. Recently, we have developed a novel machine learning model to predict drug response based on all TCGA and CCLE data. This unique deep learning machine first compresses data into essential representations of high-dimensional genomic data, and then a predictor neural network precisely predicts drug’s inhibitory concentration for the tumor: higher inhibitory concentration indicates drug resistance or sensitive to drug vice versa. Based on our preliminary result, we propose to expand our study with a new deep learning model termed “Supervised Adversarially Learned Inference” (SALI) (Aim 1) and a unique “transfer learning” module to transfer learned knowledge from cancer cell-lines to patient tumor cells (Aim 2), a major obstacle in translational research to bring in vitro experiment result into actual tumor samples. While this generative adversarial network model allows us to capture genomic knowledge more precisely, it also enable us to design several innovative methods to study drug resistance or resistance to drug combination. The successful completion of the project will provide us a novel deep learning model to predict the efficacy of therapeutic drugs, enable a new mode to prioritize cancer treatment, accelerate the discovery of novel agents, and inspire future integration of AI techniques into translational research.

Read Less