Beyond Annotation: Leveraging Raw RNA‑seq Reads via
<p dir="ltr">The poster presents a new foundation-model approach that bypasses traditional RNA-seq preprocessing and instead learns directly from raw reads to detect early-stage cancers across multiple tissue types. The work reframes early detection as a sequence-level inference prob...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Published: |
2025
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p dir="ltr">The poster presents a new foundation-model approach that bypasses traditional RNA-seq preprocessing and instead learns directly from raw reads to detect early-stage cancers across multiple tissue types. The work reframes early detection as a sequence-level inference problem rather than a gene-expression problem. Instead of relying on annotated transcripts or fixed reference genomes, the model ingests fragment-level patterns, error signatures, splicing signals, and coverage irregularities that typically get discarded during alignment.</p><p dir="ltr">The visual layout highlights three layers of innovation:</p><ol><li>an architecture trained end-to-end on billions of raw read tokens;</li><li>a representation space that captures cancer-specific perturbations invisible to conventional pipelines;</li><li>multi-cancer classifiers demonstrating improved sensitivity on early-stage disease, particularly in low-tumor-fraction settings.</li></ol><p dir="ltr">Figures show the contrast between aligned vs raw-read signal loss, the model’s latent space separating tumor vs non-tumor samples, and cross-cohort performance benchmarks. The final section outlines translational pathways—from early liquid biopsy screening to integration with cfRNA-based multi-omics signals.</p><p dir="ltr">The overall message is clear: annotated RNA-seq has reached its ceiling, and treating raw reads as a high-dimensional language unlocks a new direction for early cancer detection.</p> |
|---|