CovalentInDB: a comprehensive database facilitating the discovery of covalent inhibitors

Hongyan Du, Junbo Gao, Gaoqi Weng, Jun-Jie Ding, Xin Chai, Jinping Pang, Yu Kang, Dan Li, Dong-Sheng Cao, Tingjun Hou

TL;DR

CovalentInDB is a new, comprehensive online database that consolidates scattered experimental data on covalent inhibitors. This resource aims to accelerate the design and discovery of novel covalent drugs by providing detailed information on inhibitors, their targets, and their mechanisms of action, which is crucial given the increasing clinical success of this drug class.

What Problem Does This Paper Address?

Traditionally, covalent inhibitors were viewed with caution due to concerns about off-target effects and toxicity. However, recent clinical successes have reignited interest in this class of drugs, which can offer advantages like higher potency and longer residence times. Despite this renewed interest, a significant challenge has been the lack of a centralized, well-annotated resource for experimental data on covalent inhibitors. Existing databases often fail to specify how a ligand inhibits a target covalently or which specific residue is involved. This gap makes it difficult for researchers to rationally design new covalent inhibitors, optimize existing ones, or evaluate potential off-target risks. CovalentInDB aims to fill this void by providing a comprehensive and accessible platform for this critical information.

Methods

The CovalentInDB database was built through a rigorous data collection and processing pipeline. Scientific literature and several public databases (PubChem, ChEMBL, DrugBank, PDB, UniProt) were mined for information. Key steps included:

  • Literature Curation: Manual inspection of publications identified through keyword searches ('covalent', 'irreversible') to ensure relevance and extract detailed information.
  • Data Annotation: For each covalent inhibitor, the database annotates its structure, bioactivity, physicochemical properties, and importantly, its "warhead" (the reactive functional group). The covalent reaction mechanism and experimental verification methods are also recorded.
  • Target Annotation: Protein targets are annotated with gene information, synonyms, biological functions, and sequences. Crucially, the specific nucleophilic residue in the target that forms the covalent bond is identified.
  • Structural Data Integration: Co-crystal structures of protein-inhibitor complexes were retrieved from the PDB. For targets lacking such structures, other relevant protein structures were included to indicate binding sites.
  • Computational Property Calculation: Physicochemical properties relevant to drug-likeness (e.g., logP, logS, molecular weight) were calculated using tools like RDKit and ALOGPS.
  • Quality Assessment: A ranking system was developed to assess the reliability of the data, with "Three Stars" awarded to compounds with strong experimental validation of covalent inhibition.

The database itself is implemented using an SQLite server and a Django framework, with 3Dmol.js for interactive visualization of protein-ligand interactions.

Datasets

CovalentInDB is a newly introduced resource that consolidates data from existing sources. The primary "datasets" are:

  • Collected Literature: A vast collection of scientific publications detailing covalent inhibitors and their targets.
  • Public Databases: Data sourced from PubChem, ChEMBL, DrugBank, PDB, and UniProt.

CovalentInDB itself serves as a curated benchmark dataset for covalent inhibitors, containing:

  • 4511 covalent inhibitors
  • 280 protein targets
  • 57 different reactive warheads
  • 68 approved covalent drugs

This dataset is unique in its detailed annotation of warheads, reaction mechanisms, and specific nucleophilic residues involved in covalent binding, which differentiates it from general chemical databases.

Key Insights and Ideas

  • The Resurgence of Covalent Drugs: The paper highlights that despite historical caution, covalent drugs are making a significant comeback, with 14 approved by the FDA in the last decade, underscoring their therapeutic potential.
  • Warheads are Key: The "warhead" is identified as the critical functional group that dictates the reactivity of a covalent inhibitor, making its annotation and understanding paramount for rational design.
  • Cysteine is the Primary Target Residue: Cysteine residues are the most frequent nucleophilic targets for covalent inhibitors (53%), followed by serine (25%), indicating a strong preference in inhibitor design.
  • EGFR and Fatty-Acid Amide Hydrolase 1 are Hotspots: These two targets have the highest number of associated covalent inhibitors, suggesting significant research and development efforts in these areas.
  • Michael Acceptors Dominate Warheads: Michael acceptors are the most common warhead type, used in 45.7% of inhibitors, followed by phosphonates.
  • Bridging the Gap for In Silico Design: CovalentInDB is positioned as a vital resource for developing and validating computational tools for covalent drug design, such as covalent docking and reactivity prediction.

Key Results

  • Comprehensive Data Collection: CovalentInDB houses data for 4511 covalent inhibitors, 280 protein targets, and 57 warheads, including 68 approved drugs.
  • Detailed Annotation: The database provides extensive annotations, including structures, bioactivity, physicochemical properties, reaction mechanisms, and experimental verification methods.
  • Interactive Visualization: The web interface allows users to visualize protein-ligand interactions and highlight warhead substructures, aiding in understanding binding modes.
  • Multiple Search and Browse Functions: Users can search by text, sequence, or structure, and browse by warhead, reaction mechanism, binding site, or covalent drugs, facilitating diverse research queries.
  • Downloadable Data: All data is downloadable, enabling offline analysis and integration into other workflows.

Limitations and Future Work

  • Data Completeness: While comprehensive, the database relies on published literature, meaning some experimental data might still be scattered or unpublished.
  • Accessibility of cBinderDB: The paper notes that a previous database, cBinderDB, is no longer accessible, highlighting the importance of maintaining CovalentInDB.
  • Future Updates: The authors plan to continuously update the database every six months.
  • User Upload Functionality: A feature to allow users to upload their newly published data will be added.
  • Integration of Crystal Structures: More crystal structures of proteins bound with covalent inhibitors will be collected and integrated.
  • Systematic Study of Binding Modes: A deeper analysis of protein-ligand binding modes will be performed.
  • Covalent Site Identification Tool: A function to identify and visualize potential covalent attack sites on proteins will be developed.

Summary updated

BibTeX
@article{du2021covalentindb,
  author = {Hongyan Du and Junbo Gao and Gaoqi Weng and Jun-Jie Ding and Xin Chai and Jinping Pang and Yu Kang and Dan Li and Dong-Sheng Cao and Tingjun Hou},
  title = {{CovalentInDB: a comprehensive database facilitating the discovery of covalent inhibitors}},
  journal = {Nucleic Acids Res.},
  year = {2021},
  doi = {10.1093/nar/gkaa876},
}
Figure