Mengji Zhang
My research focuses on computational biology, with an emphasis on generative models for bulk and spatial transcriptomics, and foundation models for DNA sequences. I develop machine learning methods for omics data integration, molecular modeling, and computational neuroscience. My long-term goal is to advance the understanding of biological systems through computational algorithms.
News
Experience
Research Interests
Computational Transcriptomics
Removing unwanted variations from batch effects, platform differences, purity, and other unknown factors. Developing generative models for spatial transcriptomics (Visium and Xenium).
DNA Sequence Models
Joint modeling of cell types, microenvironment, and foundation DNA sequence models to decode regulatory logic at single-cell resolution.
Metabolomics & Structure–Property Relationships
Increasing profile reproducibility, decoding the mapping space between metabolites and measured profiles, and quantitative modeling of molecular structure–property relationships.
Computational Neuroscience
Understanding neural dynamics and the relationship between neural activities and behaviors.
Software
DeepAdapter
A universal deep neural network for eliminating unwanted variations caused by batch effects, platform differences, purity, and other unknown factors in transcriptomic studies.
pip install deepadapter
Cover Articles
Highlighted Research
Integrating Large-Scale and Heterogeneous Transcriptome Datasets
We develop DeepAdapter, a universal deep neural network to eliminate various undesirable variations from transcriptomic data, enabling accurate identification of true biological signals.
Decoding the Relationship Between Olfactory Perception and Molecular Structure
We develop Mol-PECO, a deep learning model to predict olfactory perception from molecular structures, addressing the long-standing challenge of olfactory information decoding.
Stabilizing Metabolic Profilings for Clinical Diagnosis
We report a deep stabilizer for ultra-fast, label-free mass spectrometry detection, overcoming the data quality limitations of metabolic profiling for real-time patient monitoring.
Publications
Full list on Google Scholar. † = co-first author
Gallery
