TransBind2: Improving Transcription Factor-DNA Binding Prediction with Multimodal Data and Bidirectional Cross Attention
Accurate genome-wide prediction of transcription factor (TF)-DNA binding remains challenging because many models focus mainly on DNA sequence and overlook chromatin context and TF structure. We previously developed TransBind, a protein-aware model that combines TF and DNA representations through cross-attention. Here, we introduce TransBind2, which improves on TransBind in several ways. It…
TransBind2 is an advanced model that significantly enhances the prediction of transcription factor (TF)-DNA binding, addressing the shortcomings of earlier models that primarily focused on DNA sequence and neglected chromatin context and TF structure. The earlier TransBind utilized a protein-aware approach by merging TF and DNA representations via cross-attention.
TransBind2 takes this foundation further by integrating DNase-seq accessibility and genome mappability tracks as supplementary inputs, employing a multimodal protein language model (ProstT5) to capture both TF sequence and structure, and applying bidirectional cross-attention, enabling DNA and protein features to inform and enhance each other.
Additionally, TransBind2 frames the prediction task as a binary classification of individual triplets, thereby facilitating generalization to new TFs and cell types. This improved framework was evaluated across 690 human ChIP-seq experiments involving 161 TFs and 91 cell types. The results demonstrate that TransBind2 outperforms both TransBind and other existing baselines, achieving a macro AUROC of 0.9648 and AUPR of 0.4215.
Notably, this represents an improvement of at least 12.67% in AUPR over previous models. Furthermore, a cross-species zero-shot test on mouse data showcases the model's robustness, indicating its potential for broader applicability. Saliency analysis reveals that TransBind2 can pinpoint TF-binding peaks with a median error of 12-38 base pairs, even when trained on discrete window-level labels.
Ablation studies further substantiate that TF structure, chromatin accessibility, and bidirectional attention each contribute positively to the model's performance enhancement. In summary, TransBind2 exemplifies the synergistic benefits of integrating TF structure with chromatin context, leading to more accurate and broadly applicable predictions in the field of TF-DNA binding.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
