MergeDNA: Context-aware Genome Modeling with Dynamic Tokenization through Token Merging

TL;DR

MergeDNA introduces a hierarchical, context-aware genome modeling approach that dynamically adapts tokenization to genomic sequence complexity, improving performance on DNA benchmarks and multi-omics tasks.

Contribution

It proposes a novel hierarchical architecture with dynamic tokenization and context-aware pre-training, addressing variability in genomic sequence complexity.

Findings

Outperforms existing tokenization methods on DNA benchmarks

Achieves superior results in multi-omics tasks

Effective in both fine-tuning and zero-shot settings

Abstract

Modeling genomic sequences faces two unsolved challenges: the information density varies widely across different regions, while there is no clearly defined minimum vocabulary unit. Relying on either four primitive bases or independently designed DNA tokenizers, existing approaches with naive masked language modeling pre-training often fail to adapt to the varying complexities of genomic sequences. Leveraging Token Merging techniques, this paper introduces a hierarchical architecture that jointly optimizes a dynamic genomic tokenizer and latent Transformers with context-aware pre-training tasks. As for network structures, the tokenization module automatically chunks adjacent bases into words by stacking multiple layers of the differentiable token merging blocks with local-window constraints, then a Latent Encoder captures the global context of these merged words by full-attention blocks. Symmetrically employing a Latent Decoder and a Local Decoder, MergeDNA learns with two pre-training tasks: Merged Token Reconstruction simultaneously trains the dynamic tokenization module and adaptively filters important tokens, while Adaptive Masked Token Modeling learns to predict these filtered tokens to capture informative contents. Extensive experiments show that MergeDNA achieves superior performance on three popular DNA benchmarks and several multi-omics tasks with fine-tuning or zero-shot evaluation, outperforming typical tokenization methods and large-scale DNA foundation models.