DART-Eval: A Comprehensive DNA Language Model Evaluation Benchmark on Regulatory DNA

Abstract

Recent advances in self-supervised models for natural language, vision, and protein sequences have inspired the development of large genomic DNA language models (DNALMs). These models aim to learn generalizable representations of diverse DNA elements, potentially enabling various genomic prediction, interpretation and design tasks. Despite their potential, existing benchmarks do not adequately assess the capabilities of DNALMs on key downstream applications involving an important class of non-coding DNA elements critical for regulating gene activity. In this study, we introduce DART-Eval, a suite of representative benchmarks specifically focused on regulatory DNA to evaluate model performance across zero-shot, probed, and fine-tuned scenarios against contemporary ab initio models as baselines. Our benchmarks target biologically meaningful downstream tasks such as functional sequence feature discovery, predicting cell-type specific regulatory activity, and counterfactual prediction of the impacts of genetic variants. We find that current DNALMs exhibit inconsistent performance and do not offer compelling gains over alternative baseline models for most tasks, while requiring significantly more computational resources. We discuss potentially promising modeling, data curation, and evaluation strategies for the next generation of DNALMs. Our code is available at https://github.com/kundajelab/DART-Eval.

Figures (13)

• Figure 1: An overview of DART-Eval tasks and settings
• Figure 2: Regulatory DNA syntax is sparse, combinatorial, and cell-type-dependent LungLiverDNA_Chromosomerauluseviciute_2024
• Figure 3: Distributions of zero-shot accuracies across 1443 transcription factor motifs, testing the ability to distinguish motif instances from background sequences. Vertical and horizontal lines represent 70% accuracy thresholds. In the likelihood setting, models identify most but not all motifs. In the embedding settings, models fail to distinguish motifs from background sequences.
• Figure 4: UMAP of model embeddings for sequences with experimentally identified cell-type-specific activity, colored by the true labels. The baseline embedding is a vector of canonical motif instances identified by FIMO. Numerical values are adjusted mutual information scores between true labels and a k-means clustering with $k=50$, along with a 95% confidence interval across clustering seeds, measuring the clustering quality w.r.t. the true labels. Only the baseline model yielded a useful embedding space for distinguishing sequence features for cell-type-specific activity.
• Figure S1: Correlations of per-motif embedding-based accuracies for each pair of models. Diagonals represent accuracy distribution for each model