The MAPSHB and MAPSHB-Ligand Models

Short hydrogen bond (SHB), whose donor and acceptor heavy atom separation (R) is shorter than or equal to 2.7 Å, are prevalent in proteins and protein-ligand complexes. However, the exact determination of SHBs in a protein requires the biomolecular structure to be at atomic resolution. The Machine Learning Assisted Prediction of Short Hydrogen Bonds (MAPSHB) and Machine Learning Assisted Prediction of Protein-Ligand Short Hydrogen Bonds (MAPSHB-Ligand) models allow effective predictions of SHBs from a given protein structure. The MAPSHB model is designed for SHBs that are formed between two amino acids and have the donor residue on the side chain of an amino acid. The MAPSHB-Ligand model is for SHBs that are formed from the side chain of an amino acids and a small molecule ligand.

On this page, you can upload a protein structure at any structural resolution and obtain the likely occurrence of protein-protein or protein-ligand SHBs. You will receive the results via email. These results might help the refinement and prediction of protein structures.

• The uploaded protein structure must be in the PDB or mmCIF format, and there is a size limit of 10 MB for the file. The protein structure must contain all atoms, including the hydrogen atoms in amino acid residues and small molecule ligands. The hydrogen atoms are used for the determination of hydrogen bonds. You can add them to the amino acid residues using a software of your choice prior to this analysis. For example, in ChimeraX, you can type "addh" to add hydrogens, and in AmberTools, you can use "reduce" to add hydrogens. Then you can save the structure as a PDB file.
• From the uploaded protein structures, we treat a pair A-H•••B as a hydrogen bond if the heteroatoms are O or N atoms, 2.3 Å ≤ R ≤ 3.2 Å, and the A-H-B angle ≥ 135º. It is then categorized as a SHB or normal hydrogen bond (NHB).
• The output file is in the comma-separated value (.csv) format. Each line contains the information of a hydrogen bond. Example outputs from the MAPSHB and MAPSHB-Ligand models are shown below.

MAPSHB Model

 MAPSHB-Ligand Model

• The first three columns include the residue/ligand names, residue/ligand numbers, atom names (after the @ sign) and the hydrogen bond distance from your input PDB structure. The hydrogen bond distances (R) are for your reference, and they are not used in our predictions.
• The next two columns report the probability of this hydrogen bond to form a SHB and a recommended hydrogen bond class from the MAPSHB or MAPSHB-Ligand model. The recommended hydrogen bond class is based on a classification threshold of 0.870. The models identify the hydrogen bond as a SHB if its probability is ≥ the classificatioin threshold, or a NHB if the probability is below the threshold. You can choose a different threshold to define the hydrogen bond class, as discussed on this page.

If you are interested in exploring SHBs in protein structures yourself, please visit the MAPSHB Colab notebook and MAPSHB-Ligand Colab notebook pages. 

You can access the protein structures, input files and source codes for the development of the MAPSHB and MAPSHB-Ligand models here.

If you use the MAPSHB model in your structural refinement, please cite the following reference
S. Zhou, Y. Liu, S. Wang and L. Wang, “Effective prediction of short hydrogen bonds in proteins via machine learning method”, Sci. Rep., 12, 469 (2022).

If you use the MAPSHB-Ligand model in your structural refinement, please cite the following reference
S. Zhou, Y. Liu, S. Wang and L. Wang, "Chemical Features and Machine Learning Assisted Predictions of Protein-Ligand Short Hydrogen Bonds", Sci. Rep., 13, 13741 (2023)

Upload a protein structure