Systems, Computational, and Synthetic Biology

 
Page under development

Project 1: GeneRep and nSCORE

Two major difficulties in computational biology are: (i) How to set a threshold cutoff level to maximize sensitivity while minimizing the false discovery rate?, and (ii) How to integrate ranking parameters known individually to influence network hierarchy to maximize predictive accuracy? To resolve these difficulties, we have developed a suite of tandem platforms, GeneRep and nSCORE that integrate precise and comprehensive global gene network generation with an automated node importance scoring framework with limitless sets of parameters and thus applicable to any type of networks and node statistics inputs.

(PIC of GBM GeneRep network here)

Project 2: Master Regulatory Gene Networks in Major Human Cancers

Each human cancer type possesses a unique set of genetic and epigenetic signatures, which likely explain each cancer’s disparate behaviors and response to various forms of therapy. However, most, if not all, cancer types also share many molecular and signaling similarities, irrespective of the tissue or organ of origin, suggesting there could be a common or universal regulatory gene network underpinning malignant transformation in general, which could provide the key to understanding cancer origin and developing cancer type-agnostic therapies. This project focuses on identifying such a universal signature by leveraging the predictive robustness of nSCORE and novel machine learning workflows upon the GeneRep-created reference gene networks of 28 major human cancers in the Cancer Genome Atlas (TCGA) (Make LINK HERE), followed by experimental validation in both in vitro and animal models.

Project 3: Deep Learning in Biology

Recently, advances in artificial intelligence (AI) have made tremendous breakthroughs in many areas of science. In many cases, AI systems surpass human capability in complex pattern recognition tasks, especially in medical imaging and diagnostic data such as radiographic images, EKG, histologic slides. We are developing advanced AI systems, particularly deep learning algorithms to solve complex problems in science and medicine, in close integration with our network analysis platform. Four major areas of our interests in AI in translational medical research are:

1. Precision Medicine: We are interested in developing a novel system to identify best fit drugs with the best probable response for cancer patients using integrated data: next generation DNA sequencing, RNA sequencing, imaging, histology and electronic health records. The uniqueness of our approach is a powerful combination of our innovative network analysis and deep learning algorithm.

2. Target identification: We are creating the CELL deep learning model that will integrate omics data (RNAseq, microarray expression profiles) with scientific literature using Natural Language Processing algorithm. This model would allow us to interrogate cells using computers (e.g. in silico knock down) to identify the causal gene(s) of a particular disease.

3. Acceleration of drug development: We are studying virtual drug screening using deep learning networks to identify new drugs or to repurpose existing drugs for new indications.

4. Discovery applications: We are developing AI engines to address fundamental questions in biology, such as in silico ChipSeq, predictor of binding sites of new transcription factors based on their DNA sequences, predictor of tertiary protein folding based on primary amino acid sequence.

Project 4: MCDS (Molecular Cell Diary System)

Cell fate determination is critical for cellular functions, from development, tissue repair and regeneration, to pathologic states like tissue fibrosis and cancer. Master fate determinants are poorly defined in many tissues because they are usually expressed transiently in a very small number of cells during the early stages of development, making it difficult to identify and isolate them. Even if these rare early committed cells could be isolated and potential fate factors identified, it remains a challenge to determine which factor controls which lineage since the true fate becomes harder to ascertain after the cells would have been extracted. Cancer can be viewed as a failed fate determination state and the cell origins of most cancers remain unclear. To investigate these questions, we have created MCDS, an adaptive system that will simultaneously record 1) lineage identity of individual cells (i.e. lineage tracer); 2) frequency of cellular events (i.e. cellular clock); and 3) timing and dynamics of cellular events (i.e. cellular memory/barometer). When coupled with single-cell genomics, MCDS will provide unparalleled capabilities to determine in great detail the genetic origin of each cell, timing of cellular events of interest (e.g. cell fate determinants in normal development or driver mutations during tumor progression), and molecular differences among cells (e.g. early vs. late fate committed cells) and interactions that underlie diverse potentials for different cellular outcomes.

Project 5: 3D Printing a Tumor Microenvironment

The development of novel therapies of cancer must be tested extensively in vitro before work begins preclinical animal models and eventual clinical application. The most common in vitro models for GBM are usually conducted in 2D cultures, which are known to undergo extensive phenotypic changes and exhibit markedly different responses to treatment compared to in vivo models. The need for better in vitro tumor models that can guide the development and predict the in vivo efficacy of new and effective GBM therapies is urgent. We are currently collaborating with Dr. Tommy Angelini’s group, a soft materials engineering lab whose work focuses on 3D printing physiomimetic tissue models that more closely recapitulate the 3D growth patterns and the tumor microenvironment of in vivo tumors. The primary objective of this project is to create a representative model of GBM that can then be used to study its biology in vitro and assess the efficacy of novel treatments of GBM.