MIPD: Molecules, Imagings, and Clinical Phenotype Integrated Database

Jiaojiao Zhao, Min Wu, Meihua Wan, Xue Li, Jie Li, Qin Liu, Minghao Xiong, Mengjie Tu, Jun Zhou, Shilin Li, Jie Zhang, Jiangping Fu, Yin Zhang, Chungang Zhao, Litong Qin, Xue Yang, Hong Zhao, Yan Zhang, Fanxin ZengDatabase J. Biol. Databases Curation 2025 · DOI

TL;DR

This paper introduces MIPD, a novel integrated database that links molecular data (RNA, proteins, metabolites), imaging features, and clinical phenotypes in colorectal cancer (CRC). By providing a comprehensive platform for cross-modal analysis, MIPD aims to improve our understanding of tumor heterogeneity and pave the way for more personalized treatment strategies.

What Problem Does This Paper Address?

Tumor heterogeneity is a major hurdle in cancer treatment, leading to a subset of patients who don't respond to standard therapies. While individual datasets like genomics, imaging, and clinical data offer valuable insights, integrating these diverse data types to understand their complex interrelationships has been a significant challenge. Existing databases often focus on single modalities or limited correlations, leaving a gap in a holistic view that could unlock more personalized treatment approaches. This paper addresses this gap by creating a unified database that connects molecules, imaging, and clinical phenotypes in colorectal cancer.

Methods

The core of this work is the development of the Molecules, Imaging, and Clinical Phenotype Correlation Database (MIPD). The authors collected data from 942 colorectal cancer patients, encompassing six datasets: imaging, RNA, pathway, protein, metabolite, and clinical data.

Key methodological aspects include:

  • Data Integration: MIPD integrates a vast array of data, including 9965 genes, 5449 proteins, 1121 metabolites, 283 pathways, 854 imaging features, and 73 clinical factors.
  • Imaging Feature Extraction: High-quality CT scans were used to extract 854 imaging features using a 3D Slicer plugin, covering various statistical, shape, and texture-based analyses.
  • Molecular Data Processing: RNA sequencing data was processed into transcripts per million (TPM), and genes were filtered against public databases. Protein and metabolite data were also processed and filtered for abundance.
  • Integrated Analysis: Spearman correlation analysis was employed to assess relationships between imaging features and molecular data, as well as clinical factors. Statistical tests like DESeq2 and limma were used to identify significantly different molecules across phenotypes.
  • Predictive Modeling: Machine learning models, specifically LASSO and logistic regression, were developed to predict gene expression levels from imaging features. This involved testing different pipeline strategies based on the number of correlated imaging features.
  • Database Implementation: The processed data was organized into a MySQL database, with a user-friendly, interactive web interface built using HTML/CSS, JavaScript, PHP, and MySQL.

Datasets

The study utilized data from 942 colorectal cancer patients. The datasets integrated into MIPD include:

  • Imaging Data: 591 CRC patients with CT DICOM data.
  • RNA-seq Data: 142 CRC patients.
  • Protein Data: 31 CRC patients.
  • Metabolomics Data: 56 CRC patients.
  • Clinical Data: 159 CRC patients.

Notably, 31 patients had complete multi-modal data across all categories. The paper also references publicly available databases like TCGA, TCIA, GEO, and STRING, which are linked within MIPD for further reference.

Key Insights and Ideas

  • The Power of Integration: The central idea is that integrating multi-modal data (molecules, imaging, clinical) is crucial for a deeper understanding of cancer heterogeneity and for developing personalized treatments. MIPD is presented as a significant step towards achieving this.
  • Imaging as a Window into Molecular Changes: The study highlights that imaging features can predict gene expression levels, suggesting that radiomics can serve as a non-invasive proxy for underlying molecular alterations.
  • Uncovering Cross-Modal Correlations: MIPD facilitates the discovery of complex relationships, such as how specific imaging features correlate with particular RNA, protein, or metabolite profiles, and how these, in turn, relate to clinical outcomes.
  • Personalized Treatment Potential: By identifying distinct molecules and imaging features associated with clinical phenotypes and survival, MIPD provides a foundation for stratifying patients and tailoring treatment strategies.

Key Results

  • MIPD integrates data from 9965 genes, 5449 proteins, 1121 metabolites, 283 pathways, 854 imaging features, and 73 clinical factors.
  • The database contains 497,570 correlations between imaging features, molecules, and clinical factors.
  • 1961 imaging-based models were developed to predict RNA expression levels.
  • The platform supports querying by molecular symbols, clinical factors, and imaging features, offering a user-friendly interface for exploration.
  • The database provides downloadable analysis results in various formats, including pathway matrices, imaging feature matrices, and PDF files.

Limitations and Future Work

Limitations:

  • The paper acknowledges that while the study focused on correlations between imaging features and other factors, further exploration of relationships among RNAs, proteins, metabolites, and clinical factors is warranted.

Future Work:

The authors propose several directions to enhance MIPD:

  • Integration of Additional Multi-omics Data: Incorporating genome and exome data to explore their correlations with imaging features and clinical phenotypes.
  • Intercorrelation Analyses: Developing analyses for intercorrelations among genes, RNAs, proteins, metabolites, and clinical factors.
  • Gene Module Analysis: Implementing correlation analyses between gene modules and imaging features.
  • Online Analysis Function: Developing a feature that allows users to input genes and imaging features for real-time analysis.
  • Expanded Visualization Options: Improving visualization tools to enhance data interpretation and user experience.

Summary updated

BibTeX
@article{zhao2025mipd,
  author = {Jiaojiao Zhao and Min Wu and Meihua Wan and Xue Li and Jie Li and Qin Liu and Minghao Xiong and Mengjie Tu and Jun Zhou and Shilin Li and Jie Zhang and Jiangping Fu and Yin Zhang and Chungang Zhao and Litong Qin and Xue Yang and Hong Zhao and Yan Zhang and Fanxin Zeng},
  title = {{MIPD: Molecules, Imagings, and Clinical Phenotype Integrated Database}},
  journal = {Database J. Biol. Databases Curation},
  year = {2025},
  doi = {10.1093/database/baaf029},
}
Figure