MetabOmicsmulti-omics
A comprehensive metabolism-oriented integrated multi-omics analysis method structurally designed to accommodate genomics, transcriptomics, proteomics, and metabolomics datasets. It centers on the construction of an integrated multi-omic interaction network that incorporates a wide range of biological interactions, including gene expression, translation, transcription factor activity, and post-transcriptional regulation via microRNAs. To capture the cascading effects of molecular changes, it utilizes information diffusion models, such as Linear Threshold Diffusion, to propagate fold-changes throughout the system.
Biomarker and Therapeutic Target Discovery for Alzheimer’s Disease: neurodegeneration
Managing computational analysis for a transnational EU consortium on neurodegenerative disease, concluding April 2027, supported by Biomark-X ↗ and Biomark ↗ for biomarker analysis — part of a broader push to move computational insight into clinical practice. Our goal is to discover novel non-invasive biomarkers and therapeutic targets for Alzheimer’s disease using omics and data science approaches.
Metabolitics multi-omics
An algorithm for systems-level analysis of biochemical network changes, applied to breast cancer, Crohn's disease, and colorectal cancer with over 90% average diagnostic accuracy. Filed as an international patent for low-cost, non-invasive clinical decision support; made accessible through the web-based MetaboliticsDB ↗.
Systems biology platforms infrastructure
Database-enabled tools that integrate metabolic pathway data with systems biology models for simulation, visualization, and querying — most recently extended to model genome-wide viral mutation probabilities with CovMutEx ↗.
Vertical Omics Data Integration data integration
Public metabolomics databases offer a large number of datasets. However, most datasets include measurements for only a very small fraction of the known metabolites. Hence, simply putting together these studies leads to very sparse datasets, which do not lend themselves well to training machine learning models. We developed novel approaches for vertical dataset merging across datasets. Variational autoencoders (VAE) are heavily employed for imputation model training. The developed approaches can readily be applied to other multi-omics integration challenges.
Metabolic Flux Interval Prediction ML
Flux Variability Analysis (FVA) is the gold standard for computing reaction flux intervals. However, its reliance on linear programming makes it computationally intensive, often requiring hours or days for large cohorts on complex genome-scale metabolic network models. mFLIP is a machine learning-based framework for predicting flux intervals across metabolic pathways. mFLIP provides a fast and accurate alternative to traditional FVA-based approaches for metabolic flux interval estimation. By leveraging supervised learning on FVA-derived data, it enables scalable analysis of large cohorts while maintaining high predictive performance, making it a practical tool for large-scale metabolic studies.
Drug Repositioning ML
The main aim of this project is to develop a metabolism-based pipeline that will allow a computational investigation of whether the currently approved drugs for any disease are suitable for use in treating diseases other than their intended target. In particular, we develop tools to simulate the effect of each drug on patient metabolism, and then assess whether the drug can push a diseased metabolic profile back toward a healthy one. This approach can also surface entirely new drug targets that are not on the market.
Readmission Risk Prediction medical informatics
Readmission of patients after a short period of time from their discharge costs billions of dollars to governments every year. This research aims to develop an integrated readmission risk monitoring system that will continuously monitor patients, and act as an auxiliary decision support system to provide clinicians with the risk of readmission during the entire period of a patient's stay, as well as after the discharge through the periodic remote measurements taken by the patient at home.
Identifying Missed Opportunities in Chronic Diseasesmedical informatics
Chronic diseases such as diabetes and heart failure often suffer from late diagnoses, which can lead to preventable complications, inefficient care management, and increased healthcare costs. This project presents a timeline-aware machine learning framework that estimates disease probability at each hospital visit per-patient and compares these predictions with actual diagnosis date to identify missed opportunities.
Computational Patient medical informatics
Static record-keeping in the form of electronic health records (EHRs) is increasingly inadequate in an era defined by multimodal data and foundation models. Computational Patient is a continuously evolving, general-purpose digital representation that integrates heterogeneous inputs, including EHRs, wearable device data, and multi-omics profiles, into a unified latent state-space. Unlike task-specific digital twins, this framework supports cyclic feedback loops, uncertainty quantification, and simulation-based reasoning.
Cancer Risk Prediction genomics, clinical informatics
Cancer susceptibility is driven by a complex interplay of genetic predispositions and environmental exposures, yet traditional analyses frequently evaluate these factors in isolation. We investigate the combinatorial impact of targeted Single Nucleotide Polymorphisms (SNPs), clinical demographics, and behavioral risks on lung and colorectal cancer etiology.
Signatures of Biological Information Flow multi-omics, network analysis, systems biology
Understanding how genomic variability is translated into stable or dysfunctional metabolic phenotypes remains a central challenge in systems biology. We develop a multi-omic, information-theoretic framework that maps transcriptomic variation onto metabolic network activity using deterministic Gene-Protein-Reaction (GPR) rules applied to the Recon3D human metabolic network. By integrating paired transcriptomic and metabolomic data across cancer and neurodegenerative cohorts, we quantify pathway-level entropy, variance buffering (canalization), and topological information flow.
Risk Stratification of Lymphedema medical informatics
Breast cancer–related lymphedema (BCRL) is a frequent and clinically significant complication of breast cancer treatment. Early identification of high-risk patients may facilitate personalized surveillance and preventive interventions. We develop an interpretable machine learning framework to enable accurate
BCRL risk stratification and support personalized surveillance and survivorship care in breast cancer patients. We also designed a web-based clinical decision support system.
Causal Probabilistic Clinical State Framework medical informatics, AI
Clinical artificial intelligence (AI) systems are predominantly trained under the assumption that each patient corresponds to a single, well-defined ground truth label. However, increasing evidence from diagnostic variability, noisy label learning, and uncertainty quantification suggests that medical labels are inherently probabilistic, fluid, and context-dependent. We develop a Causal Probabilistic Clinical State Framework (C-PCSF). In this paradigm, disease is modeled not as a discrete label, but as a posterior distribution over a latent physiological manifold, coupled with intervention-aware causal dynamics and utility-driven decision policies.
Metabolic Deconvolution of Bulk Omics systems biology, multi-omics
Single-cell RNA sequencing provides high-resolution characterization of the tumor microenvironment, yet its limited scalability, high cost, and technical variability restrict widespread clinical adoption. Consequently, bulk tissue multi-omics remains the dominant profiling strategy in clinical oncology, despite its inherent limitation in averaging heterogeneous cellular signals and obscuring cell-type–specific states. We develop a topology-guided representation learning framework that integrates paired bulk transcriptomic and metabolomic data to construct high-resolution pseudo-single-cell–like representations.
Fluxomics Meets Polygenic Risk Scores systems biology, multi-omics
Complex genetic diseases result from intricate interactions among genetic variants and environmental factors. While genome-wide association studies (GWAS) have identified numerous risk loci, their limited mechanistic insight underscores the need for integrative, systems-level approaches. We develop a novel computational framework that integrates polygenic risk scores with genome-scale metabolic models to contextualize genetic susceptibility within a condition-specific metabolic landscape.
