Multiscale Modeling as a Service (MMaaS) - Research Areas

Type 1 Diabetes (T1D)

Type 1 diabetes  is a chronic autoimmune disease that can develop at any age, in which the pancreas is not producing enough insulin needed for regulating blood glucose levels. Despite years of  research, to date, the exact cause of T1D is still unknown and there is no cure for the disease. By collaborating with Johns Hopkins University, we have identified a unique diabetogenic peptide in a novel immunogenic pancreatic cell line. This work is fundamental in that it discovers the putative immunodominant peptide which causes T1D. We are currently expanding this work to de novo design of peptides for potential therapeutic uses by utilizing molecular biophysical modeling to study the immune system’s recognition of the identified disease-causing antigen as well as proposed antigen variants for therapeutic purposes. 

Related publications:

Cancer Immunotherapy

Immune checkpoint blockade (ICB) therapies that based on inhibitors targeting cytotoxic T lymphocyte antigen 4 (CTLA-4) and programmed cell death protein 1 (PD-1) or its ligand (PDL-1) have significantly improved the survival of some patients with metastatic cancers. However, response rates to ICB therapies remains a challenge as it rarely exceeded 40%. One of the main causes is due to the diversity in the genes encoding the human leukocyte antigen (HLA) system and the T cell receptor (TCR). Our very recent study in collaboration with the Memorial Sloan Kettering Cancer Center showed that patients with certain HLA genotypes (such as HLA-B44) respond much better than those with other HLA genotypes (such as HLA-B15). Also, the loss of HLA heterozygosity and less mutational load are correlated with decrease of the ICB response of the patient. Based on these newly found results, we continue to explore  how the genetic variation affects the interaction between TCR, HLA, and the antigen using combination of  experimental  and computational approaches with deep learning techniques, aiming at improving the future development of cancer immunotherapy.

Related publications:

Huntington’s Disease (HD)

Huntington’s disease is a neurodegenerative disease that stems from the aggregation of Huntingtin proteins when the Glutamine-expanded sequence (Q length) is longer than 36. However, it has not been established why long Q lengths lead to the disease due to a previous dearth of structural information. Our group, in collaboration with the Cure for Huntington's Disease Initiative (CHDI), was the first in the world to generate a comprehensive structural map for the pathogenic fragment of the human Huntingtin protein with various Q lengths. With the advantage in our ownership of this unique comprehensive dataset, we will explore the pathological and regulatory networks of HD and expand upon our knowledge for potential therapeutic application.

Related publications:

Model-enhanced analytics for treatment and diagnostics of cardiac diseases 

Cardiovascular disease remains a leading cause of death in United States. Biophysical models of cardiovascular system and heart function are promising analytic tools to assist medical practice and clinical trials. Over the years, we have developed detailed multi-scale models of cardiac mechanics that can leverage data collected across several scales of physiology in clinical and pre-clinical settings. Some examples include data from in vitro experiments, multi-modality cardiac imaging, and other cardiac functional measurements. The team has significant expertise in development of detailed 3D high resolution finite element models of cardiac mechanics capable of providing personalized insights into the state of health of patients with heart disease. To make feasible model optimization of cardiovascular system, we have recently developed a novel multi-fidelity strategy for model order reduction of 3-D finite element models of ventricular mechanics. The significant improvement in computational efficiency of the low order models over the finite element model counterparts allows for efficient parametric explorations and solutions of inverse optimization problems, which is crucial for the application of biophysical models in clinical settings. Our group continues to extend, validate and improve our computational models with an aim to develop a highly sophisticated cardiac modeling platform that assimilates multimodal data and provides patient-specific diagnosis and treatment plans. We are exploring potential application of our developed pipeline for providing improved interpretation and early assessment of pharmacological therapies (e.g., inotropic drugs, medication used to treat hypertension etc.) in presence of multiple confounding factors. Currently, the platform is extended to provide enriched analysis of low fidelity cardiac imaging data (e.g., echocardiography) and enhanced predictions of surgical procedure outcomes. Our other research directions are techniques for coupling of machine learning (ML) and biophysical models to make ML models more interpretable.

Related publications:

Model-enhanced prediction of cardiotoxic risk 

Cardiac toxicity is an adverse effect presented by many drugs, including the ones that are not directly targeted to alter heart function. Cardiac safety remains a leading cause for discontinuation of drugs at different phases of drug development. In the past, several FDA-approved drugs were pulled from the market for causing arrhythmias (disturbances in electrical activity of the heart) that resulted in sudden cardiac death in certain patients. Other drugs can inadvertently change the myocardial contractility impairing cardiac function. Model-based risk prediction promises improved interpretation of underlying pathophysiology and mechanism-based classification of risk. The team has developed, over the years, multi-scale models of cardiac electrophysiology and mechanics that are highly suitable for evaluation of drug-effects and extraction of risk markers. We have also recently incorporated sensitivity analysis and uncertainty quantification approaches to our computational pipeline, which could be used for quantification and improvement of the confidence in the predicted risk and to explore the intrinsic structure of model derived metrics. We continue to extend, validate and improve our computational models with the aim to develop a sophisticated framework for improved risk predictions of novel drugs in the early stages of drug development pipeline and risk evaluations across different patient populations.

Related publications:

Neuronal disease progression modeling 

Electrophysiological recordings can capture states of populations of neurons in a pre-clinical settings at different time points, disorder states, or combinations of both. The phenotypes of these cells emerge from complex interactions of intracellular processes. Understanding how these intracellular processes can be modulated to transition a neuron from the phenotype of one condition to the phenotype of another condition can provide a template for relevant therapeutics in the conditions of interest. Using a population of models parameter search and analysis framework developed at IBM research, we are able to generate parameter sets for mechanistic models that span the ranges of electrophysiological properties (neuronal phenotypes) within and between disease burden and natural aging categories. We have applied these techniques to understanding phenotypic progression in the population of striatal spiny projection neurons in Huntington's Disease.

Understanding polypharmic effects within single neurons and neuron populations 

Pre-clinical Huntington's Disease (HD) research has demonstrated that striatal spiny projection neurons (SPNs) are dysfunctionally hyper-excitable in multiple rodent models of HD. Researchers from the University of Pittsburgh's Drug Discovery Institute (DDI) recently demonstrated the application of high-content screening (HCS) and quantitative systems pharmacology methods, to hey were able to identify synergistic combinations of small molecule probes that revealed a convergence of pathways implicating activation of the PKA (Protein Kinase A) network in enhanced neuroprotection in HD. In collaboration with researchers from the DDI, we are connecting HCS measures with models of intracellular pathways leading to key downstream targets of the PKA network, such as DARPP-32 phosphorylation, to simultaneously regulate multiple membrane currents that coherently modulate cell excitability. We can use an understanding of this regulation within SPNs to predict the perturbations to electrophysiological phenotype resulting from drug application.

Protein-protein interaction screening (PPI) 

The large magnitude of protein-protein interaction (PPI) pairs within the human interactome necessitates the development of predictive models and screening tools to better understand this fundamental molecular communication for critical diseases such as cancer, HIV/AIDS, Huntington's Disease (HD), influenza, and Alzheimer's Disease (AD). However, despite enormous efforts from various groups to develop predictive techniques in the last decade, PPI complex structures are in general still very challenging to predict due to the large number of degrees of freedom. In this work, we explore the idea of applying AI/ML, free energy perturbation (FEP) calculations, conventional molecular dynamics (MD) simulations, and steered molecular dynamics (SMD) simulations to enhance the PPI protein docking scores for better ranking of all predicted structures using state-of-the-art molecular modeling techniques.

Related publications:

Natural sweeteners screening

Obesity is a serious health problem which can lead to comorbidities such as cardiovascular diseases, type II diabetes, high blood pressure, depression, osteoarthritis and various cancers. Recently, many scientific studies have shown clear link between excessive sugar intake and obesity. Therefore, more and more health conscious consumers are putting great emphasis on natural and low calorie products these days. In this project, we apply structural biology and molecular modeling techniques to construct the whole structure of human sweetener receptor for systematically studying of the interactions among the sweet heterodimer receptor, sweeteners and modulators using combined methods of molecular dynamics simulations and deep learning for  the discovery of new natural sweeteners with zero- or low calorie.

Related publications:

  • Jose Manuel Perez-Aguilar, Seung-gu Kang, Leili Zhang, and Ruhong Zhou,
    Modeling and Structural Characterization of the Sweet Taste Receptor Heterodimer,
    (submitted to ACS Chem Neuro)