Summary
BioKERN is a multimodal spatial representation-learning framework that incorporates biological structure as an explicit, learnable inductive bias for histology-to-transcriptomics tasks. Instead of emphasizing only instance-level cross-modal matching between paired histology and transcriptomics spots, BioKERN constructs a training-time biological kernel that combines transcriptomic similarity with spatial proximity, then uses this kernel to provide graded neighborhood supervision and to regularize the embedding geometry. Evaluation is conducted using a fixed, model-independent biological neighborhood definition shared across all compared methods. Across the Mouse Brain Visium and Human Liver GSE240429 datasets, BioKERN consistently improves biological neighborhood retrieval in both single-scale and multi-scale settings. The framework was developed by Seungik Cho and Betul Orcan-Ekmekci and is described in arXiv paper 2608.24823. This work addresses a limitation of prior histology-transcriptomics objectives, which may underutilize cases where non-paired spots share molecular or spatial context.
Overview
Field: Machine Learning
Authors: Seungik Cho, Betul Orcan-Ekmekci
Published: 2026-08-25
arXiv: 2608.24823
Summary
Spatially resolved biology requires representations that preserve biological neighborhood structure rather than only exact cross-modal correspondences. Existing histology-transcriptomics objectives can emphasize instance-level matching even when non-paired spots share molecular or spatial context.
BioKERN is a multimodal spatial representation-learning framework that incorporates biological structure as an explicit, learnable inductive bias. The key ideas:
- Training-time biological kernel: constructed by combining transcriptomic similarity and spatial proximity.
- Graded neighborhood supervision: the kernel is used to provide graded neighborhood-level supervision rather than only pairwise instance matching.
- Embedding regularization: the kernel regularizes the geometry of learned embeddings to preserve biological neighborhood structure.
- Model-independent evaluation: evaluation uses a fixed, model-independent biological neighborhood definition shared by all methods.
Results
Across the Mouse Brain Visium and Human Liver GSE240429 datasets, BioKERN consistently improves biological neighborhood retrieval in both single-scale and multi-scale settings.
Links
- Paper: https://arxiv.org/abs/2608.24823
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178634095