Particle.news

Download on the App Store

Johns Hopkins Publishes MIGHT, Improving AI Reliability for Liquid Biopsy Detection

The publicly released models quantify uncertainty to curb false positives by incorporating inflammatory-disease data.

Overview

  • Two PNAS studies introduce MIGHT and CoMIGHT, designed for high‑dimensional, small‑sample datasets, with code released at treeple.ai for independent testing.
  • Across 1,000 blood samples, aneuploidy features yielded 72% sensitivity at 98% specificity, outperforming other AI approaches in consistency and accuracy.
  • A companion study found ccfDNA fragmentation signatures in autoimmune and vascular diseases and associated them with elevated inflammatory biomarkers.
  • Incorporating inflammatory‑disease data into MIGHT reduced non‑cancer false positives, though inflammation remains a confounder and clinical validation is still required.
  • CoMIGHT’s multi‑signal analysis of 125 early breast, 125 early pancreatic, and 500 control samples indicated potential gains for some early detections, and a related editorial flagged eight barriers to clinical integration.