EvoLen: Evolution-Guided Tokenization for DNA Language Model

arXiv cs.LG / 4/13/2026

💬 OpinionIdeas & Deep AnalysisModels & Research

共有:

Key Points

EvoLen proposes evolution-guided tokenization for DNA language models, arguing that DNA token boundaries should be driven by functional motifs preserved under evolutionary constraint rather than linguistic-like regularities.
The method incorporates cross-species evolutionary signals by stratifying/grouping sequences, training separate BPE tokenizers per group, and merging vocabularies with rules that prioritize preserved patterns.
EvoLen further applies length-aware decoding using dynamic programming to better maintain motif-scale functional units during representation.
In controlled experiments, EvoLen improves preservation of functional sequence patterns, differentiates genomic contexts, and better aligns with evolutionary constraint while matching or outperforming standard BPE on DNALM benchmarks.
The work concludes that tokenization choice acts as a critical inductive bias for DNALM performance and interpretability, and that evolutionary information yields more biologically meaningful token representations.

Abstract

Tokens serve as the basic units of representation in DNA language models (DNALMs), yet their design remains underexplored. Unlike natural language, DNA lacks inherent token boundaries or predefined compositional rules, making tokenization a fundamental modeling decision rather than a naturally specified one. While existing approaches like byte-pair encoding (BPE) excel at capturing token structures that reflect human-generated linguistic regularities, DNA is organized by biological function and evolutionary constraint rather than linguistic convention. We argue that DNA tokenization should prioritize functional sequence patterns like regulatory motifs-short, recurring segments under evolutionary constraint and typically preserved across species. We incorporate evolutionary information directly into the tokenization process through EvoLen, a tokenizer that combines evolutionary stratification with length-aware decoding to better preserve motif-scale functional sequence units. EvoLen uses cross-species evolutionary signals to group DNA sequences, trains separate BPE tokenizers on each group, merges the resulting vocabularies via a rule prioritizing preserved patterns, and applies length-aware decoding with dynamic programming. Through controlled experiments, EvoLen improves the preservation of functional sequence patterns, differentiation across genomic contexts, and alignment with evolutionary constraint, while matching or outperforming standard BPE across diverse DNALM benchmarks. These results demonstrate that tokenization introduces a critical inductive bias and that incorporating evolutionary information yields more biologically meaningful and interpretable sequence representations.