Research Projects

January, 2025- Present

Supervised Contrastive Learning for interpretable and robust Molecular Pattern Recognition in High-dimensional Spatial Biomedical Data.

Developed an AI-driven supervised contrastive learning framework for imaging mass spectrometry and spatial transcriptomics data that enables the discovery of sex/BMI/age-invariant and demographically robust molecular biomarkers. The model identifies FTU (functional tissue unit)-specific low-dimensional representations that enhance tissue classification, spatial segmentation, and downstream biological analysis. This approach improves the reliability of biomarker identification, supports precision diagnostics, and provides molecular insights through biologically grounded feature attribution and integrated SHAP-based explainability.

March 2024 - December 2025 

Integrating ion-mobility and Imaging Mass Spectrometry for Chemistry-Aware Dimension Reduction. 

Developed a computational framework that integrates m/z information, spatial ion images, and drift time bin (ion-mobility) to analyze high-dimensional imaging mass spectrometry datasets. The approach yields an isotope-aware, low-dimensional representation that preserves meaningful chemical structure, facilitating more accurate isotope candidate detection, enhanced data interpretability, and a more reliable downstream molecular analysis. This work supports biomarker discovery and translational biomedical research, and a manuscript based on this study is currently under review with other co-authors.

April,2023 -- February, 2024

Computational Methods for Robust and Interpretable Domain-aware Dimension Reduction in Imaging Mass Spectrometry Data.

Led the design of a chemistry-informed dimensionality reduction method to improve automatic detection of isotope candidates in high-dimensional spectral biomedical datasets. The method identifies an isotope-aware low-dimensional representation that preserves significant chemical and structural information, which enhances both the accuracy and efficiency of molecular analysis, enabling the identification of key molecular markers relevant to drug discovery. This work has resulted in a manuscript currently under review.

January,2016 - October, 2022

Sensing Matrix Optimization and Feature Extraction for Image Processing.

• Developed a robust sensing matrix based on an equiangular tight frame, reducing pairwise correlation and ensuring tightness. This advancement enhances compressive sensing in applications like wireless communication, medical imaging, satellite imaging, radar systems, and signal processing.

• Proposed a representation-based feature extraction model to address challenges of insufficient inter-class margin and loss of manifold structure. This approach is particularly effective in classification and recognition tasks with limited training data.

• A regression-based feature extraction model is developed to preserve the underlying structure of the data with sufficient inter-class margins.