Lactate Metabolism Gene Signature and H&E Image Model for HNSCC

59 lactate-related genes form a signature linked to tumor proliferation and immune suppression across TCGA, GEO, and single-cell RNA-seq datasets. A deep learning model predicts lactate metabolic states from routine H&E whole-slide images, validated in 12 TCGA cancer types and an independent HNSCC cohort. The work integrates multi-omics and digital pathology to propose a digital biomarker.

Use Cases

• Validate the 59-gene lactate signature against immune infiltration metrics in TCGA or GEO datasets.
• Apply the deep learning framework to H&E whole-slide images to classify tumors as LAC_H or LAC_L states.
• Correlate model-predicted lactate activity with immunohistochemistry results for LDHA and MCT1 protein expression.
• Analyze associations between high lactate activity (LAC_H) and therapeutic resistance outcomes in HNSCC patient cohorts.

Strengths

• Defined a specific 59-gene lactate signature for biological characterization.
• Model validation across 12 TCGA cancer types with AUC performance of 0.78–0.89.
• External biological validation in an independent SAZHU-HNSCC cohort using IHC.

Limitations

• Primary data is a 44.1 KB DOCX file describing the study, not the underlying genomic or image datasets.
• Access to the actual multi-omics, single-cell, or H&E image data used requires sourcing from external repositories like TCGA and GEO.
• The document's utility is methodological; applying the signature or model requires users to obtain and process separate, large-scale data.

Provenance

Source

figshare, authored by Bohai Feng.

Collection Method

Integrative analysis of public datasets (TCGA, GEO, single-cell RNA-seq) and development of a computational deep learning framework.

Time Range

null

Freshness

Last updated March 2026.

Geography

null