QSP Technology Development

QSP Technologies


Computational Pathology

Tissue diagnosis is a mix of science and art because disease biology is heterogeneous and human visual interpretation is subjective, yet both are key sources of information driving QSP. To support the pathologist’s cognitive burden and minimize the human error, we have two projects in visual learning and tissue pattern recognition from histopathology images:

Canonical pointwise mutual information maps depicting various forms of spatial intratumor heterogeneity.
  1. Learning the origins of cancer tissue (e.g., breast, kidney, lung, melanoma, etc.) from its spatial architecture using deep neural networks for classifying tissue origins with performance comparable to that of expert pathologists;
  2. Segmenting and triaging regions of interest for diagnostics using both deep and shallow learning to classify histological structures and rank cancer risk.

The Computational Pathology program is a collaboration with Chakra Chennubhotla, PhD, Associate Professor, Department of Computational and Systems Biology and CEO, Co-Founder of SpIntellx™ (a computational and systems pathology company) along with UPMC clinicians in cancer and liver diseases, including Adrian Lee, PhD (breast cancer), John Kirkwood, MD and Hassane Zarour, MD (melanoma), and Satdarshan (Paul) Singh Monga, MD and Jaideep (Jai) Behari, MD, PhD (hepatocellular carcinoma and NAFLD).


Machine Learning           

Machine learning is a key component of QSP that is used to infer pathways

QuartataWeb is an integrated chemogenomics server for searching and inferring: ligands and pathways for a certain target; targets and pathways for a certain drug or chemical; or perform enrichment analysis for a list of drugs/targets/drug-drug combinations.

of disease progression based on gene expression profiles from “normal” and disease tissue samples. The selected pathways allow identification of known molecular targets that serve as candidate targets for pharmacological and/or genomic perturbation to investigate disease mechanism. Machine learning tools such as QuartataWeb (http://quartata.csb.pitt.edu) are also used to predict drug- or chemical-target interactions from databases including DrugBank and STITCH in order to identify a focused library of disease mechanism “probes” that includes known and predicted drugs to test in MPS experimental models.

Application of Machine Learning tools in QSP is a UPDDI collaboration with Ivet Bahar, PhD (Chair of the Dept. of Computational and Systems Biology), Timothy Lezon, PhD (Computational and Systems Biology), Chakra Chennubhotla, PhD (Computational and Systems Biology) along with UPMC clinicians including Erin Kershaw, MD (Chief of the Division of Endocrinology and Metabolism, Dept. of Medicine), Jaideep (Jai) Behari, MD, PhD (Director of the UPMC “FLOW” Center (NAFLD Clinic), Dept. of Medicine), Vijay Yechoor, MD (Division of Endocrinology and Metabolism, Dept. of Medicine), and Ramon Bataller, MD, PhD (Division of Gastroenterology, Hepatology and Nutrition (AFLD Program)) to evaluate the clinical relevance of selected pathways, targets and drugs.

Human Microphysiology Systems (MPS) & Database (MPS-Db)

The application of human MPS in QSP has grown out of the recognition that

Integrated MPS allow dynamic real-time measurements of biosensor cells in normal and disease states. Evolving the models toward all human cell types derived from a single iPSC line. All data is uploaded to the MPS-Db; analyses and modeling conducted within the QSP framework.

animal models and simple 2D monocultures of cells do not reflect the complexity and specificity of human physiology, toxicology, and disease mechanisms. MPS are experimental models that use patient-derived primary cells or tissue-resident adult stem cells (AdSCs), embryonic stem cells (ESCs), or induced pluripotent stem cells (iPSCs), to recapitulate enough tissue/organ functions to serve as useful models in the drug discovery and development pipeline. There is the added potential to create a personalized platform for preclinical trials using these patient-derived cells. MPS experimental models being developed, tested and applied include human liver models of toxicity and disease (D. Lansing Taylor, PhD, UPDDI), a pancreatic islet model of type 2 diabetes (Ipsita Banerjee, PhD, Dept. of Chem/Petroleum Engineering), an adipose tissue model of metabolic syndrome (Lauren Kokai, PhDDepartment of Plastic Surgery and Rosalyn Abbot, PhD, CMU), and a liver niche model of metastatic melanoma (Mark Miedel, PhD, UPDDI).

The MPS-Db (https://mps.csb.pitt.edu) was designed and developed in the UPDDI to capture, manage and analyze data, and share complex studies performed using MPS. It provides tools for quantifying reproducibility, creating computational models, PBPK / PD for in vitro in vivo extrapolation (IVIVE), and for comparing experimental data with preclinical, clinical and post-marketing data to support QSP. It is currently being used for evaluation of performance and reproducibility among organ models tested in the NCATS Tissue Chips Testing Centers (TCTC) program where Pitt is the analytical center.

The development and application of the MPS models involves UPDDI and a collaborative network of clinicians and engineers from: the UPMC Hillman Cancer Center (John Kirkwood, MD, Hassane Zarour, MD); Vanderbilt University (John Wikswo, PhD); University of Wisconsin (William Murphy, PhD); UPitt Dept of Pathology and the McGowan Institute of Regenerative Medicine (Alex Soto-Gutierrez, MD, PhD)

Computational Modeling

A central element of QSP is the iterative computational/experimental

Detailed computational model of immunoreceptor signaling mediated by the high-affinity receptor for IgE (Fc epsilon R1). 

feedback loop. In order to understand the biological mechanisms of disease onset and progression, it is necessary to formalize certain aspects of the experimental system into a mathematical model that can be manipulated in silico, such as agent-based models (ABMs) or systems of ordinary differential equations (ODEs). In the context of the computational model, the difference between “diseased” and “healthy” states arises from changes in parameters, such as reaction rates or concentrations. Changes in the computational model that promote the disease phenotype indicate hypothetical mechanisms of disease progression. MPS models are used to test these hypotheses and identify drugs that rectify these changes. If that reverses the disease state, then the computational model has successfully identified a disease mechanism; if not, then the computational model is refined, a new hypothesis is generated and then experimentally tested. Development and application of computational models is a collaboration between the UPDDI, Timothy Lezon, PhD (Computational and Systems Biology) and Jim Faeder, PhD (Computational and Systems Biology).

Computational Models are also used to predict pharmacokinetics (PK). Understanding the PK properties of a candidate drug is important in determining proper administration of the compound to achieve therapeutic benefit. Human organotypic MPS models, singly or when coupled, have great potential as a platform for early PK studies. We are implementing physiologically based pharmacokinetic (PBPK) prediction tools in the MPS-Db. The implementation and application of PK modeling is a collaboration between Mark Schurdak, PhD (UPDDI), Lawrence Vernetti, PhD (UPDDI), Tong Ying Shun, PhD (UPDDI), Jan Beumer, PhD (Pharmaceutical Sciences).


The PHDA’s University of Pittsburgh spinout company introduces spatial intelligence and explainable artificial intelligence to improve pathology and biopharma research.

Read more