Genetics

bioMoR: Biology-Guided Mixture-of-Recursions for Effective Genomic Learning

Koushik Howlader, Tirtho Roy, Md Tauhidul Islam, Wei Le

Featured August 15, 2026

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

This paper introduces a smart way for AI models to learn from genes and pathways by using biological maps to decide which parts of the data need more thinking, making the learning faster and more accurate.

In depth

The paper introduces bioMoR, a framework that integrates biological knowledge into Mixture-of-Recursions (MoR) Transformer models for genomic data. It addresses the inefficiency of standard Transformers by adaptively allocating computation to biologically relevant tokens. This is achieved by smoothing token embeddings, biasing self-attention, and guiding the routing mechanism using gene co-expression or pathway relationships, leading to improved performance and efficiency.

Key Takeaways

1
bioMoR is the first framework to apply Mixture-of-Recursions to gene-level and pathway-level genomic learning, enabling adaptive computation.
2
The framework integrates structured biological knowledge at three key sites: embedding smoothing, attention biasing, and adaptive routing, enhancing model effectiveness and interpretability.
3
The proposed method significantly improves predictive performance (e.g., 8.2 percentage points in macro-F1) while substantially reducing computational cost (75% fewer parameters, up to 58% fewer FLOPs) compared to biology-agnostic Transformers.

Conceptual Flow

HIGH LEVEL

Methodology (The "Logic")

The model takes gene data, uses biological maps to make sense of it, and then decides how much to "think" about each gene, focusing more on important ones.

Gene Data

Biological Maps

Smartly Process

Better Predictions

Results (The "Impact")

By using biological maps, the new method makes much better predictions with less computer power than older methods that treat all genes the same.

Old Way: Slow, Less Accurate

New Way: Fast, More Accurate

Compare Performance

Big Improvement

This breakdown was generated by SciGrove. Get the same analysis — intuition, storyboard, peer review, a runnable prototype and a glossary — on any paper you upload or paste a DOI for.