Alignment-free metal ion-binding site prediction from protein sequence through pretrained language model and multi-task learning

Abstract

More than one-third of the proteins contain metal ions in the Protein Data Bank. Correct identification of metal ion-binding residues is important for understanding protein functions and designing novel drugs. Due to the small size and high versatility of metal ions, it remains challenging to computationally predict their binding sites from protein sequence. Existing sequence-based methods are of low accuracy due to the lack of structural information, and time-consuming owing to the usage of multi-sequence alignment. Here, we propose LMetalSite, an alignment-free sequence-based predictor for binding sites of the four most frequently seen metal ions (Zn2+, Ca2+, Mg2+ and Mn2+). LMetalSite leverages the pretrained language model to rapidly generate informative sequence representations and employs transformer to capture long-range dependencies. Multi-task learning is adopted to compensate for the scarcity of training data and capture the intrinsic similarities between different metal ions. LMetalSite was shown to surpass state-of-the-art structure-based methods by more than 19.7%, 14.4%, 36.8%, and 12.6% in AUPR on the four independent tests, respectively. Further analyses indicated that the self-attention modules are effective to learn the structural contexts of residues from protein sequence.

Prediction of Ca2+binding site in proteins with a fast and accurate method based on statistical mechanics and analysis of crystal structures

“…Most of the current prediction methods are data-driven, which hinge on the extraction of features from both amino acid sequences and structures to facilitate the training of predictive models 12 . Some approaches centred around sequences [13][14] , and others exhibit a structural emphasis akin to methods like Fold-X 15 , MIB [16][17] , and BioMetAll 18 . Other methodologies take a dual approach, combining sequence-based and structural information to achieve a more comprehensive analytical standpoint [19][20] .…”

Predicting the effects of mutations on protein solubility using graph convolution network and protein language model representation

“…For example, transfer learning technique has shown ability to transfer knowledge from related tasks with large‐scale data to the target task with limited data 28 . Our previous works have highlighted the power of pretrained protein language model ProtTrans 29 on metal ion‐binding site prediction 30 …”

“…28 Our previous works have highlighted the power of pretrained protein language model ProtTrans 29 on metal ion-binding site prediction. 30 Another limitation of existing sequence-based solubility change prediction is the lack of protein structure information. Deep learning methods have been shown to benefit from protein structure information for protein solubility prediction, 31 stability changes prediction, 32 protein design, 33 protein-protein interaction, 34 and drug discovery.…”