Transcription factors are proteins that regulate the expression of genes by binding to specific genomic regions known as Transcription Factor Binding Sites (TFBSs), typically located in the promoter regions of those genes. Accurate prediction of these binding sites is essential for understanding the complex gene regulatory networks underlying various cellular functions. In this regard, many deep learning models have been developed for such prediction, but there is still scope of improvement. In this work, we have developed a deep learning model which uses pre-trained DNABERT, a Convolutional Neural Network (CNN) module, a Modified Convolutional Block Attention Module (MCBAM), a Multi-Scale Convolutions with Attention (MSCA) module and an output module. The pre-trained DNABERT is used for sequence embedding, thereby capturing the long-term dependencies in the DNA sequences while the CNN, MCBAM and MSCA modules are useful in extracting higher-order local features. TFBS-Finder is trained and tested on 165 ENCODE ChIP-seq datasets. We have also performed ablation studies as well as cross-cell line validations and comparisons with other models. The experimental results show the superiority of the proposed method in predicting TFBSs compared to the existing methodologies. The codes and the relevant datasets are publicly available at https://github.com/NimishaGhosh/TFBS-Finder/.

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 $28$ distinct datasets across $7$ tasks, with input lengths ranging from $70$ to $1000$. 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 $56 \times$ less GPU time in pre-training.