Skip to main content Skip to secondary navigation

Healing the humanity with AI and science

Main content start

1. Research Overview

 Hypothesis driven, Real World Big data Empowered, Translational AI Analytics

  •  Population health R&D: Collaborating with multiple state Health Information Exchanges (HIE), my lab uses tens of millions of real-world state-wide EMRs, accessing more than 60M population, to develop risk surveillance systems that forecast aspects like high impact disease progression, resource utilization, and mortality across a diverse patient demographic with different social determinants. This prompts timely clinical actions by simplifying intervention orders and crafting care strategies tailored to address modifiable patient risk components.
  • First-in-class molecular diagnostics: My lab has developed unique LCMS based multi-omic approaches that allow the simultaneous absolute quantification of thousands of metabolites and proteins in blood and FFPE pathological slides to predict clinical outcomes. Our collaborations with key opinion leaders in pregnancy disorder and pediatric diseases, such as Kawasaki disease, preeclampsia and preterm birth, have been productive leading to first-in-class molecular diagnostics. For example, we participated in the NIH Technology Accelerator Challenge for Maternal Health (Co-PI: Ling) and won US National Semi-Finalist 07/2022.
  • Computer-aided pathology (CAP) and computer-aided medical imaging analytics (CAMIA): My lab has developed deep learning-based computational solutions to decode clinical outcome-correlating signals in pathological whole slide images and echocardiograms. Our multi-modality and multi-omics approaches synergize to promise the next generation of disease diagnostics and risk stratification solutions.

2. Research Objectives

2.1. Use of deep learning approaches to develop predictive models for chronic diseases and resource utilizations to improve global population health for longevity.

Over the past decade, my laboratory has pioneered the use of electronic medical records (EMRs) to develop deep learning algorithms for patient risk prediction. These algorithms have been integrated into collaborating US Health Information Exchange (HIE) platforms, including Maine State (HIN) and New York City (Healthix), to deliver real-time insights to providers. By predicting the risk of disease, cost, overall mortality, and future events, such as hospital admissions, readmissions, emergency visits, suicide, overdose, and stroke, our algorithms enable healthcare providers to identify populations in need of focused interventions and administer preventative care earlier. We use machine learning to continuously improve risk prediction accuracy and natural language processing to rapidly interpret new data and make calculated predictions with minimal labor input. Our laboratory has 20+ peer-reviewed publications on disease and risk prediction, supporting HIE performance results and providing a layer of transparency unmatched within the population health industry.

Nippon Life, the largest insurance company in Japan, and the Translational Medicine Laboratory (PI Ling) will collaborate to develop deep learning models for chronic diseases and improve population health. The $3 million two-year contract is expected to be signed in November 2023. In this study, we hypothesize that complex diseases progress through three states: normal, pre-disease (or critical transition), and disease. Japan's vast trove of real-world EMR and wearable data can be mined to identify critical transition states of high-impact chronic diseases. Deep learning can play a key role in population health by reducing hospitalizations, especially for chronic diseases, and predicting future costs for healthcare providers to make better business and clinical decisions. Our rationale is that identification of the clinical mechanism(s) underlying chronic disease functional deterioration advance our knowledge of chronic disease progression and risk prediction and healthcare resource utilization forecasting using breakthrough AI deep learning technology and EMR and wearable device combined health big data.  The delivery of risk engine goals is expected to significantly translate the medicine of global healthcare management.

2.2. Development of neuroblastoma tissue diagnostic utility through deep learning-based image analytics and targeted multiplex proteomics.

Neuroblastoma (NB) is a prevalent and aggressive extracranial childhood cancer. The accurate determination of differentiation grading and evaluation of clinical outcomes are essential in guiding treatment plans for this disease. The objective of our research team is to develop an innovative diagnostic utility that can effectively guide treatment decisions for NB. This tool will utilize deep learning-based H&E pathological image analytics to accurately grade NB, while also incorporating targeted multiplex FFPE proteomics for prognostic purposes. Our clinical team, led by co-PI Dr. Shimada, has played a pivotal role in developing the International Neuroblastoma Pathology Classification System (the Shimada system). This system identifies specific morphological characteristics to determine differentiation grading and evaluate NB clinical outcomes. Through our pilot efforts, we successfully employed convolutional neural networks (CNNs) for differentiating NB tumors from hematoxylin and eosin (H&E) stained histology sections. This advancement contributed to the precision of NB computer-aided pathology (CAP). We also developed an innovative CAP/CLIA clinical workflow that utilizes precise laser capture microdissection (LCM) to isolate targeted cells from the tumor or tumor microenvironment population. This technique utilizes formalin-fixed paraffin-embedded (FFPE) tissue and enables targeted multiplex proteomics analyses. Additionally, our team has designed and synthesized internal standard (IS) reagents, enabling us to construct a proprietary cancer protein atlas with over 1200 tumor and TME protein targets using high-resolution mass spectrometry.

Our research team is strongly positioned to successfully address critical challenges in neuroblastoma, particularly in the areas of tissue grading, classification, and risk stratification. Our efforts aim to guide precise therapy for combating high-risk neuroblastoma. This three-year proposal will be funded by FDA with $4.5M (11/2023 kick off), involving significant advancements in new products and methods.

2.3. Integrated trajectories of the maternal blood/urine metabolome and proteome to predict pregnancy disorders including preeclampsia and spontaneous preterm birth.

Multifactorial diseases typically emerge with poorly defined etiologic mechanism and pathogenesis. For the past decade, my laboratory has leveraged high resolution LCMS/MS data sets through an analytic platform utilizing either in house proteomics/metabolomics or from a common repository to discover molecular disease fingerprints. My lab has employed a comprehensive unbiased multi-’omics’ approach, integrating genomics, metabolomics, and proteomics to define a molecular phenotype to diagnose disease and assess trajectory.

Given the high rates of morbidity and mortality associated with prematurity and low birth weight caused by preterm birth or preeclampsia, the ability to identify high-risk pregnancies and infants early is crucial. My lab has developed LCMS metabolomics/peptidomics/proteomics using maternal blood or urine to identify biomarkers predictive preterm birth and preeclampsia. Previously, my laboratory developed the metabolic or proteomic “clock” to determine gestational age during human pregnancy. We propose now to use the pregnancy clock as a baseline to predict pregnancy disorders.  We hypothesize that abnormal maternal metabolic/proteomic status during pregnancy negatively impacts growth and development of the fetus and leads to: a) preterm birth; b) preeclampsia; c) in utero growth restriction and small-for-gestational-age (SGA) newborns; d) metabolic vulnerability in the newborn that quantitatively reflects term, preterm, and growth restricted newborns. Furthermore, we hypothesize that the premature or SGA infants’ abnormal metabolic capacity strongly influences the onset of disease during early childhood (stunting, neurodevelopmental impairment). We foresee that these studies will serve as the foundation for follow-on studies that will seek to determine (1) the association between maternal gestational and after birth infant metabolic/proteomic markers (phenotype) in growth restricted and preterm infants; and (2) the risk of acquired neonatal diseases, specifically stunting and neurodevelopmental impairment as examples.