Knowledge-guided multi-modality transformer for multi-label genetic mutation prediction

Gexin Huang

doi:10.1016/j.patcog.2026.113047

Knowledge-guided multi-modality transformer for multi-label genetic mutation prediction

2026-01-16

0

PRE

AI

DOI：10.1016/j.patcog.2026.113047

Original

Original request for help

Save

Abstract

En 中文

• The first work to propose a multi-label framework for predicting genetic mutations from whole slide images (WSIs). • Intrinsic biological knowledge is incorporated to capture gene dependencies, enabling the model to reason about mutations in a multi-label context. • A graph-based gene encoder is proposed to capture linguistic, phenotypic, and pathway-level associations. • A label decoder is proposed to align gene priors with spatial WSI features and employs a comparative multi-label loss to improve discrimination across mutation statuses. • The framework improves accuracy and interpretability, and is validated on 9 TCGA cancer types.

Knowledge-guided multi-modality transformer for multi-label genetic mutation prediction

Abstract

Journal

Organization