Editorial: Data mining and statistical methods for knowledge discovery in diseases based on multimodal omics, volume II

Tao Wang, Miguel E. Rentería, Zhen Tian, Jiajie PengFrontiers in Genetics 2023 · DOI · OpenAlex

Abstract

This editorial article introduces a curated collection of eight research papers that push the boundaries of multi-omics data integration. By moving beyond traditional statistical constraints, these studies employ advanced machine learning, spatial modeling, and network analysis to uncover novel biomarkers and therapeutic targets for complex conditions like multiple sclerosis, sepsis, and various cancers.

Introduction

The biological "big data" revolution has provided researchers with a wealth of information—from bulk genomics to single-cell and spatial transcriptomics. However, the field faces a significant bottleneck: traditional statistical methods, such as transcriptome-wide association studies or standard Mendelian Randomization, are often limited by pre-defined modalities. These legacy approaches frequently fail to capture the full complexity of how different biological layers (DNA, RNA, proteins, etc.) interact. This Research Topic addresses the urgent need for more flexible, integrative computational frameworks that can synthesize these diverse data streams to discover actionable knowledge about human diseases.

Methodological Innovations

The papers in this volume introduce several novel methodologies designed to handle high-dimensional biological data:

  • SpatialMap: A computational tool that uses generalized linear spatial models and conditional autoregressive priors to infer gene expression profiles in spatial transcriptomics, even for genes not directly measured.
  • Network Representation & Deep Learning: Employed for predicting disease-associated microRNAs (miRNAs) by modeling biological entities as complex networks.
  • Ensemble Machine Learning: One study combined CatBoost with SHAP (Shapley Additive explanations) to create an interpretable model for predicting sepsis risk, ensuring clinicians can understand the logic behind predictions.
  • Algorithmic Clustering: The use of random walk algorithms and Monte Carlo methods to identify molecular modules that evolve across different stages of lung adenocarcinoma.
  • Two-Sample Mendelian Randomization (MR): Applied to explore causal links between lifestyle factors and specific disease outcomes.

Data Integration and Analysis

The research utilized a wide array of high-dimensional datasets to validate these methods:

  • Proteomic profiles from the serum and cerebrospinal fluid (CSF) of multiple sclerosis patients.
  • Single-cell RNA sequencing (scRNA-seq) paired with image-based transcriptomics.
  • Pan-cancer datasets (including BRCA/breast cancer) to analyze the STAT gene family.
  • SARS-CoV-2 protein sequences for phosphorylation site prediction.
  • Transcriptomic cohorts of European descent for sepsis and skin disease studies.

Key Insights and Results

The integration of these diverse data streams has led to several significant findings:

  • Spatial Contextualization: The introduction of SpatialMap highlights a shift toward "filling in the blanks," leveraging spatial correlation to predict molecular activity in tissue areas where data was previously missing.
  • Clinical Interpretability: The use of SHAP values in sepsis prediction models demonstrates the necessity of "Explainable AI" (XAI) for clinical translation, ensuring high accuracy is paired with transparency.
  • Dynamic Disease Progression: Research identified 12 core molecular modules that evolve across stages of lung adenocarcinoma, suggesting a roadmap for stage-specific treatment rather than viewing cancer as a static state.
  • Biomarker Identification:
    • Multiple Sclerosis: Identified 73 differentially expressed proteins in CSF and 22 in serum, pinpointing three specific proteins (MMP2, C8G, and CFH) as high-potential biomarkers.
  • Cancer Genomics: Pan-cancer analysis revealed the STAT gene family as a universal biomarker for predicting prognosis and guiding immunotherapy.
  • Viral Mechanisms: Successful prediction of phosphorylation sites in SARS-CoV-2 aids in understanding how the virus hijacks host cellular machinery.

Conclusion and Future Work

While these studies represent significant strides, the field remains challenged by the sheer heterogeneity of omics datasets. Future research must focus on moving from feature integration to truly holistic models, capturing the dynamic nature of molecular networks as diseases progress, and bridging the gap between computational discoveries and bedside therapeutic strategies.

Reference: Editorial article Front. Genet., 24 August 2023 Sec. Statistical Genetics and Methodology Volume 14 - 2023 | https://doi.org/10.3389/fgene.2023.1270862

Summary updated

Abstract

EDITORIAL article Front. Genet., 24 August 2023Sec. Statistical Genetics and Methodology Volume 14 - 2023 | https://doi.org/10.3389/fgene.2023.1270862
BibTeX
@inproceedings{wang2023editorial,
  author = {Tao Wang and Miguel E. Rentería and Zhen Tian and Jiajie Peng},
  title = {{Editorial: Data mining and statistical methods for knowledge discovery in diseases based on multimodal omics, volume II}},
  booktitle = {Frontiers in Genetics},
  year = {2023},
  doi = {10.3389/fgene.2023.1270862},
}

Loading related content...

Figure