We propose the first unified adversarial attack benchmark for Genomic Foundation Models (GFMs), named GERM. Unlike existing GFM benchmarks, GERM offers the first comprehensive evaluation framework to systematically assess the vulnerability of GFMs to adversarial attacks. Methodologically, we evaluate the adversarial robustness of five state-of-the-art GFMs using four widely adopted attack algorithms and three defense strategies. Importantly, our benchmark provides an accessible and comprehensive framework to analyze GFM vulnerabilities with respect to model architecture, quantization schemes, and training datasets. Empirically, transformer-based models exhibit greater robustness to adversarial perturbations compared to HyenaDNA, highlighting the impact of architectural design on vulnerability. Moreover, adversarial attacks frequently target biologically significant genomic regions, suggesting that these models effectively capture meaningful sequence features.
Splice sites play a crucial role in gene expression, and accurate prediction of these sites in DNA sequences is essential for diagnosing and treating genetic disorders. We address the challenge of splice site prediction by introducing DeepDeCode, an attention-based deep learning sequence model to capture the long-term dependencies in the nucleotides in DNA sequences. We further propose using visualization techniques for accurate identification of sequence motifs, which enhance the interpretability and trustworthiness of DeepDeCode. We compare DeepDeCode to other state-of-the-art methods for splice site prediction and demonstrate its accuracy, explainability and efficiency. Given the results of our methodology, we expect that it can used for healthcare applications to reason about genomic processes and be extended to discover new splice sites and genomic regulatory elements.
Decoding the linguistic intricacies of the genome is a crucial problem in biology, and pre-trained foundational models such as DNABERT and Nucleotide Transformer have made significant strides in this area. Existing works have largely hinged on k-mer, fixed-length permutations of A, T, C, and G, as the token of the genome language due to its simplicity. However, we argue that the computation and sample inefficiencies introduced by k-mer tokenization are primary obstacles in developing large genome foundational models. We provide conceptual and empirical insights into genome tokenization, building on which we propose to replace k-mer tokenization with Byte Pair Encoding (BPE), a statistics-based data compression algorithm that constructs tokens by iteratively merging the most frequent co-occurring genome segment in the corpus. We demonstrate that BPE not only overcomes the limitations of k-mer tokenization but also benefits from the computational efficiency of non-overlapping tokenization. Based on these insights, we introduce DNABERT-2, a refined genome foundation model that adapts an efficient tokenizer and employs multiple strategies to overcome input length constraints, reduce time and memory expenditure, and enhance model capability. Furthermore, we identify the absence of a comprehensive and standardized benchmark for genome understanding as another significant impediment to fair comparative analysis. In response, we propose the Genome Understanding Evaluation (GUE), a comprehensive multi-species genome classification dataset that amalgamates $36$ distinct datasets across $9$ tasks, with input lengths ranging from $70$ to $10000$. Through comprehensive experiments on the GUE benchmark, we demonstrate that DNABERT-2 achieves comparable performance to the state-of-the-art model with $21 \times$ fewer parameters and approximately $92 \times$ less GPU time in pre-training.
Genome sequencing technology has improved significantly in few last years and resulted in abundance genetic data. Artificial intelligence has been employed to analyze genetic data in response to its sheer size and variability. Gene prediction on single DNA has been conducted using various deep learning architectures to discover splice sites and therefore discover intron and exon region. Recent predictions are carried out with models trained on sequence with fixed splice site location which eliminates possibility of multiple splice sites existence in single sequence. This paper proposes sequential labelling to predict splice sites regardless their position in sequence. Sequential labelling is carried out on DNA to determine intron and exon region and thus discover splice sites. Sequential labelling models used are based on pretrained DNABERT-3 which has been trained on human genome. Both fine-tuning and feature-based approach are tested. Proposed model is benchmarked against latest sequential labelling model designed for mutation type and location prediction. While achieving high F1 scores on validation data, both baseline and proposed model perform poorly on test data. Error and test results analysis reveal that model experience overfitting and therefore, model is deemed not suitable for splice site prediction.