ProLIF: a library to encode molecular interactions as fingerprints.
Bouysset C, Fiorucci S. · 2021
Read ProLIF: a library to encode molecular interactions as fingerprints. by Bouysset C, Fiorucci S. (2021) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc8466659.txt.
Introduction
Bouysset and Fiorucci present ProLIF, a Python library that turns molecular contacts into fingerprints. Such fingerprints sum up how molecules touch in three dimensions. Drug projects use them for screening and machine learning. Good tools to build them were still needed.
ProLIF reads complexes from simulations, lab structures, and docking runs. It handles mixes of ligand, protein, DNA, or RNA parts. Users can retune the built-in contact types or add their own. Tutorials cover the common use cases step by step. Code snippets show how to pair the library with other analysis tools. ProLIF is free to install from the team's GitHub repository.
Catalog record
| CREATOR | Bouysset C, Fiorucci S. |
|---|---|
| YEAR | 2021 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
ProLIF: a library to encode molecular interactions as fingerprints By Cédric Bouysset, Sébastien Fiorucci Published in: Journal of Cheminformatics DOI: 10.1186/s13321-021-00548-6 ABSTRACT Interaction fingerprints are vector representations that summarize the three-dimensional nature of interactions in molecular complexes, typically formed between a protein and a ligand. This kind of encoding has found many applications in drug-discovery projects, from structure-based virtual-screening to machine-learning. Here, we present ProLIF, a Python library designed to generate interaction fingerprints for molecular complexes extracted from molecular dynamics trajectories, experimental structures, and docking simulations. It can handle complexes formed of any combination of ligand, protein, DNA, or RNA molecules. The available interaction types can be fully reparametrized or extended by user-defined ones. Several tutorials that cover typical use-case scenarios are available, and the documentation is accompanied with code snippets showcasing the integration with other data-analysis libraries for a more seamless user-experience. The library can be freely installed from our GitHub repository (https://github.com/chemosim-lab/ProLIF). Supplementary Information The online version contains supplementary material available at 10.1186/s13321-021-00548-6. INTRODUCTION Interactions between and within molecular structures are the driving force behind biological processes, from protein folding to molecular recognition. The decomposition of interactions by residues in biomolecular complexes can provide insights into structure–function relationships, and characterizing the nature of each of these interactions can guide medicinal chemists in structure-based drug discovery projects [1]. Approaches to encode the interactions observed in 3D structural data in the form of a binary fingerprint have been developed in the past [2–6] and applied successfully to a variety of projects. For example, de Graaf et al. [7] used the Tanimoto similarity between the interaction fingerprint (IFP) of a crystallographic reference and the IFP of docking poses to rescore virtual screening results on a G protein-coupled receptor (GPCR). Rodríguez-Pérez et al. [8] showed that IFPs can achieve superior predictive performance than ligand fingerprints (ECFP4) for the classification of kinase inhibitor binding modes with machine-learning models. Finally, Mpamhanga et al. [9] showed that one can use the IFP for clustering, and then shortlist a reasonable number of binding modes prior to visual inspection. More recently, the approach was also implemented for molecular dynamics (MD) simulations to study ligand unbinding [10]. While the typical IFP usually encodes pre-established interactions (hydrogen bond, π-stacking…etc.) on a per-residue basis, other implementations exist. Sato et al. [11] developed a pharmacophore-based IFP which relies on the pharmacophoric features of the ligand atoms in contact with the protein and the distance between each of these pharmacophores to generate a bitvector. Da et al. [12] developed an IFP that relies on the atomic environment of both the protein and ligand interacting atoms to set the positions of a bit in the fingerprint, rather than relying on protein residues and predefined interactions, which has the advantage of implicitly encoding every possible type of interaction. This protocol was later reimplemented in Python by Wójcikowski et al. [13], but other more classical Python-based IFP implementations exist [14–19]. In this paper, we introduce a new Python library, ProLIF, that overcomes several limitations encountered by these programs, namely working exclusively with the output of specific docking programs, not being compatible with the analysis of MD trajectories, being restricted to a specific kind of complex (usually protein–ligand complexes), depending on residue or atom type naming conventions, or not being extensible or configurable regarding interactions (Table 1).Table 1Comparative table of features available in non-commercial IFP softwareSoftwareComplexesInput formatMDConfigurableExtensibleWebCLIRef.ProLIFAllMDAnalysis and RDKitc✓✓✓IChemLPMOL2✓✓[4]PLIPAllaPDB✓✓✓[15, 18]ArpeggioAllaPDB✓✓✓[16]PyPLIF HIPPOSLPaPDBQT, MOL2✓✓[17]getContactsAlla,bVMDc✓✓[19]MD-IFPLPaMDAnalysisc✓[10]ODDTLPOpenBabel and RDKitc✓[20]MD native support of MD trajectories; Web available through a webserver; CLI command-line interface; LP Ligand–Protein; All all combinations between ligand, protein, DNA, and RNA moleculesaRelies on residue and atom naming convention to assign charges and/or aromaticitybOnly supports hydrogen-bond, hydrophobic and van der Waals contacts for LP complexescCompatible with the input formats supported by the aforementioned libraries IMPLEMENTATION ProLIF can deal with RDKit [21] molecules or MDAnalysis [22] Universe objects as input, which allows supporting most 3D molecular formats, from docking to MD simulations. While most MD topology files do not keep explicit information about bond orders and formal charges, MDAnalysis is able to infer this information if all hydrogen atoms are explicit in the structure while converting the structure to an RDKit molecule. The RDKit parent molecule is then automatically fragmented in child residue molecules based on residues name, number, and chain to make it easier to work on a per-residue basis when encoding the interactions. When calculating an interaction fingerprint, each interaction is typically defined as two groups of atoms that satisfy geometrical constraints based on distances and/or angles (Table 2). Here the selection of atoms is made using SMARTS queries (Table 3), which is more precise than relying on elements or atomic weights and is also more universal than relying on force-field-specific atom types.Table 2Interactions currently available in ProLIFInteractionLigandaProteinaDistance (Å)Angle (deg)AnionicAnionCation≤ 4.5CationicCationAnionCationPiCationAromatic(+)-ctd ≤ 4.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{n}, \overrightarrow{ctd\cdots (+)}\rangle \in \left[0, 30\right]$$\end{document}⟨n→,ctd⋯(+)→⟩∈0,30PiCationAromaticCationPiStackingAromaticAromaticctd–ctd ≤ 6.0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{n}, \overrightarrow{n}\rangle \in \left[0, 90\right]$$\end{document}⟨n→,n→⟩∈0,90min ≤ 3.8EdgeToFaceAromaticAromaticctd–ctd ≤ 6.0\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{n}, \overrightarrow{n}\rangle \in \left[50, 90\right]$$\end{document}⟨n→,n→⟩∈50,90Min ≤ 3.8FaceToFaceAromaticAromaticctd–ctd ≤ 4.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{n}, \overrightarrow{n}\rangle \in \left[0, 40\right]$$\end{document}⟨n→,n→⟩∈0,40min ≤ 3.8HBAcceptorHBAcceptorHBDonorD–A ≤ 3.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{HD}, \overrightarrow{HA}\rangle \in \left[130, 180\right]$$\end{document}⟨HD→,HA→⟩∈130,180HBDonorHBDonorHBAcceptorXBAcceptorXBAcceptorXBDonorX–A ≤ 3.5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{XD}, \overrightarrow{XA}\rangle \in \left[130, 180\right]$$\end{document}⟨XD→,XA→⟩∈130,180XBDonorXBDonorXBAcceptor\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\langle \overrightarrow{AX}, \overrightarrow{AR}\rangle \in \left[80, 140\right]$$\end{document}⟨AX→,AR→⟩∈80,140MetalAcceptorLigandMetal≤ 2.8MetalDonorMetalLigandHydrophobicHydrophobicHydrophobic≤ 4.5(−) anion; (+) cation; ctd centroid of the aromatic ring; min minimum value in the distance matrix between both aromatic rings; n normal to the aromatic ring plane; D hydrogen/halogen bond donor; A hydrogen/halogen bond acceptor; H hydrogen atom; X halogen atom; R atom linked to a halogen bond acceptoraAlthough “ligand” and “protein” are used here, all the listed interactions can be applied to any molecular complex (protein–protein, DNA–protein…etc.)Table 3SMARTS patterns used in the definition of interactionsNameSMARTS pattern(s)Anion[−{1−}]Cation[+{1−}]Aromatica1:a:a:a:a:a:1a1:a:a:a:a:1HBAcceptor[N,O,F,−{1−};! + {1−}]HBDonor[#7,#8,#16][H]XBAcceptor[#7,#8,P,S,Se,Te,a;! + {1−}][*]XBDonor[#6,#7,Si,F,Cl,Br,I]-[Cl,Br,I,At]Metal[Ca,Cd,Co,Cu,Fe,Mg,Mn,Ni,Zn]Ligand[O,N,−{1−};! + {1−}]Hydrophobic[#6,#16,F,Cl,Br,I,At; + 0] The library is designed so that users can easily modify existing interactions, as there is usually no consensus on the empirical thresholds (distance, angles) that should be used. For example, the hydrogen bond DH…A can be defined as a distance between H and A lower or equal to 3.0 Å [9] or as a distance between D and A lower or equal to 3.5 [4, 14, 20] or 4.1 Å [15, 23], and the angles constraints can also vary. ProLIF is also designed to let users define custom interactions. Each interaction is written as a Python class that implements a “detect” method which takes two RDKit molecules as input, typically a ligand and a protein residue, and outputs a Boolean (True if the interaction is present, else False) as well as the indices of atoms responsible for the interaction. All interaction classes are then gathered inside a “Fingerprint” class that can generate a bitvector from two RDKit molecules, and optionally return the atom indices. By default, the Fingerprint class is configured to generate a bitvector with the following interactions: hydrophobic, π-stacking, π-cation and cation-π, anionic and cationic, and H-bond donor and acceptor, although more specific interactions are available (see Table 2). This Fingerprint class