← BioTransfer GEO Dataset Finder
GEO series

Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions [HTAN cohort]

GSE320381 Homo sapiens (human) RNA-seq (gene expression) 145 samples Submitted 2026/05/27 Platform GPL30173Platform GPL18573
At a glance

RNA-seq (gene expression) · human · 145 samples

What it is
RNA-seq (gene expression)Gene activity measured by sequencing. The summary does not mention single-cell work, so this is most likely bulk: one average per sample, the usual starting point for comparing groups.
Organism
Homo sapiens (human)
Topic
Lung cancerThis series is part of our curated briefing on it.
Samples
145 samples: 31 conditions, mostly in duplicateGroups are read from the sample names, not from the paper. Check the list below.
Data you can get
A processed table (series matrix) and supplementary files from NCBI; the raw sequencing reads from ENA. Which of these hold counts depends on the study, so check the file list.
Paper
Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesionsGenome medicine · 2026
Summary (written by the submitters)
Background: Molecular and cellular alterations to the normal pseudostratified columnar bronchial epithelium results in the development of bronchial premalignant lesions representing a spectrum of histology from normal to hyperplasia, metaplasia, dysplasia (mild, moderate, and severe), carcinoma in situ and invasive carcinoma. Several studies have identified molecular alterations associated with lesion histology and progression.
Read the full summary ↓
The broad and continuous spectrum of histologic and molecular changes makes reproducible stratification of lesions across multiple studies challenging. Methods: We proposed a transformer-based framework that flexibly utilizes transcriptomic and histologic patterns to distinguish lesions with bronchial dysplasia or worse from normal, hyperplasia, and metaplasia. We leveraged H&E whole slide images (WSIs) of endobronchial biopsies and bulk gene expression data (GE) derived from endobronchial biopsies and brushings from previously published studies and on-going lung precancer atlas efforts obtained from patients at high-risk for lung cancer. Results: On an external testing dataset of WSIs, the model trained on WSIs plus GE achieved an area under the ROC curve (AUROC) of 0.884±0.040 compared to 0.829±0.046 for the model trained on WSIs alone. On external testing datasets of GE, the model trained on WSIs plus GE achieved an AUROC of 0.857±0.033 versus 0.713±0.098 for a model trained on GE alone. Based on these results, we leveraged data across 4 studies to train a flexible fusion model that allows one or both data modalities (WSIs and GE) to be used in training. The model achieved an AUROC of 0.906±0.034 on external testing WSIs data and 0.870±0.023 on external testing GE data. Despite model training on a binary label, model probabilities were associated with histologic grade and the model identified gene expression alterations associated with bronchial dysplasia across multiple studies. Conclusions: Our multimodal transformer outperformed models trained on a single data modality and enabled the inclusion of samples with one or both modalities during training and/or testing. It increases the flexibility, scalability, and real-world applicability of disease severity assessment that better risk stratifies bronchial premalignant lesions even when only routine histology data is accessible.
Published in
Attention-based deep learning for analysis of pathology images and gene expression data in lung squamous premalignant lesions
Xu L, Kefella Y, Zhang Y et al. · Genome medicine 2026 · PMID 41952176 · doi:10.1186/s13073-026-01636-8
This dataset
Open GSE320381 in finder → Find more human RNA-seq datasets → Lung cancer landscape →
Download · which file do I need?

Direct links to NCBI and ENA, no account and no request form. Pick by what you want to do: take the small processed table to look, the supplementary files to re-analyse, or the original raw data to start over.

1 · Just have a look

Processed table

The series matrix: one table of values per sample as the submitters processed it. The smallest download, and it opens in Excel or R. Start here to see what the data look like.

Series matrix ↓
2 · Re-analyse it

Supplementary files

The count tables and other files the authors uploaded. Take only the samples you need, or the whole study as GSE320381_RAW.tar (it can run to tens of gigabytes).

All supplementary files ↗ Whole study · RAW.tar ↓
3 · Start from the raw reads

Raw sequencing reads

The original FASTQ files, straight from the sequencer, held at ENA. Large, and you align them yourself, so only choose this if you need to re-process the study from scratch.

Raw reads on ENA ↗

What the files usually look like: For RNA-seq the counts are usually a table among the supplementary files (look for 'counts', 'expression', 'FPKM' or 'TPM' in the file name). The series matrix of a sequencing study is sometimes nearly empty.

Also filed as BioProject PRJNA1427693. Searching any of these in the dataset finder brings you back here.

NCBI GEO page ↗ Paper (PMID 41952176) ↗
Study design
31 conditions, mostly in duplicate
PCGA02_20152 ×12 PCGA02_20119 ×9 PCGA02_10417 ×8 PCGA02_20145 ×8 PCGA02_10401 ×7 PCGA02_10404 ×7 PCGA02_10420 ×7 PCGA02_20132 ×7 +31 more

Supports a between-group comparison across 137 samples.

How we read this: 31 replicated groups read from 145 sample titles; they account for 137 of them. These groups are guessed from the sample names, not taken from the paper, so check them against the sample list below before relying on them.

Samples in this study

+ 105 more — browse all 145 samples with per-sample file links →

More Lung cancer datasets

Search all human RNA-seq datasets in GEO →

Share this dataset

Metadata from NCBI GEO, cached and refreshed periodically — the NCBI page above is authoritative. Downloads link straight to NCBI/ENA; nothing is proxied through BioTransfer.