bio-covalent-design

Designs covalent inhibitors and warheads targeting cysteine (most common, 98% of covalent drugs), lysine, serine, threonine, tyrosine, and aspartate residues, with explicit handling of warhead reactiv

SKILL.md

GitHub

Version Compatibility

Reference examples tested with: RDKit 2024.09+, OpenEye / AutoDock Vina 1.2+ (for covalent extensions), GOLD (commercial), DOCKovalent (web service), HCovDock 1.0+.

Before using code patterns, verify installed versions match. If versions differ:

  • Python: pip show rdkit then help(rdkit.Chem) to check signatures
  • CLI: check version output of each docking tool

If code throws ImportError, AttributeError, or TypeError, introspect the installed package and adapt the example to match the actual API rather than retrying.

Covalent Inhibitor Design

Design molecules that form covalent bonds with target protein residues. The "covalent revolution" (Lonsdale & Ward 2018) made TCIs (Targeted Covalent Inhibitors) clinically validated: KRAS G12C inhibitors (sotorasib, adagrasib), BTK inhibitors (ibrutinib), and EGFR inhibitors (osimertinib) are recent successes. Postdoc-grade covalent design requires balancing intrinsic reactivity (must form bond) vs selectivity (only the intended residue), reversibility (irreversible vs reversible covalent), and drug-likeness (warheads can hurt PK).

For warhead substructure filtering (in non-covalent contexts), see chemoinformatics/substructure-search. For non-covalent docking, see chemoinformatics/virtual-screening. For pose validation, see chemoinformatics/pose-validation.

Reactive Residue Taxonomy

Residue % of covalent drugs Reactivity Notes
Cysteine ~98% High (nucleophile thiol) Most accessible; preferred
Lysine ~1% Moderate (amine) Less reactive; selective for sulfonyl fluoride
Serine <1% Low (alcohol, requires activation) β-lactam, boronate
Threonine very rare Low Boronate, aldehyde
Tyrosine very rare Moderate (phenol) Sulfonyl fluoride, fluorosulfate
Aspartate/Glutamate very rare Low (carboxylate) Aldehyde Schiff base

Cysteine is the dominant target because:

  • Soft nucleophile (matches soft electrophiles)
  • Low background reactivity (rare in proteins, ~1.7%)
  • Distinguishable from common nucleophiles (GSH, off-target Cys)

Warhead Chemistry

Warhead SMARTS pattern Reactivity Reversibility Cys-selective
Acrylamide C(=O)C=C Moderate (Michael acceptor) Irreversible Yes
Chloroacetamide C(=O)CCl High (SN2) Irreversible Yes
α-haloketone [CX3](=O)C[F,Cl,Br] Very high Irreversible Yes (but reactive)
Vinyl sulfone S(=O)(=O)C=C Moderate (Michael) Irreversible Yes
Sulfonyl fluoride S(=O)(=O)F Moderate Irreversible Lys/Tyr/Ser
Fluorosulfate (SuFEx) OS(=O)(=O)F Moderate Irreversible Tyr/Lys
Aldehyde C(=O)[H] Variable Reversible (covalent equilibrium) Cys/Lys/Ser
Boronate (B-OH or B(OH)2) B(O)O Moderate Reversible Ser/Thr
Nitrile C#N Low Reversible (Cys-S adduct) Cys
Epoxide C1OC1 High Irreversible Cys/Lys/Asp
α,β-unsaturated ketone [CX3](=O)C=C Moderate (Michael) Irreversible Cys
Isothiocyanate N=C=S High Irreversible Cys/Lys
Maleimide C(=O)N(C(=O))C=C Very high Irreversible Cys
Cysteine-selective heterocycle various Moderate Variable Yes (designed)

Practical hierarchy: Acrylamide is the modern default for cysteine-selective TCIs (KRAS G12C, EGFR, BTK). Chloroacetamide is more reactive (faster) but less selective.

Decision Tree by Scenario

Goal Warhead choice Reactivity tier
Cysteine TCI, drug candidate Acrylamide Moderate (~kinact/Ki ~10^3-10^5 M^-1 s^-1)
Cysteine probe (chemical biology) Chloroacetamide High (kinact/Ki ~10^4-10^6 M^-1 s^-1)
Lysine TCI (uncommon) Sulfonyl fluoride Moderate
Tyrosine TCI Fluorosulfate (SuFEx) Moderate
Reversible covalent (KRAS G12C-like) Acrylamide with α-substitution Moderate reversibility
Activity-based protein profiling (ABPP) Iodoacetamide / chloroacetamide Very high
Boronic acid inhibitor (proteasome) Boronate Reversible
Aldehyde inhibitor (calpain) Aldehyde Reversible covalent

Kinetics: kinact / Ki

Covalent inhibition kinetics:

  • Ki: reversible binding affinity (initial, like non-covalent IC50)
  • kinact: rate of covalent bond formation (sec^-1)
  • kinact/Ki: second-order rate constant, "covalent efficiency" (M^-1 s^-1)

Modern best practice: report kinact/Ki, not just IC50. Two compounds with same IC50 can have very different kinact/Ki:

  • Low Ki, low kinact: tight binding, slow covalent bond
  • High Ki, high kinact: loose binding, fast covalent bond
kinact/Ki range (M^-1 s^-1) Interpretation Reference
> 10^5 Highly efficient covalent inhibitor Chloroacetamide probes, fragment-warhead TCIs
10^3 - 10^5 Standard for TCI; clinical candidate Sotorasib (AMG510) KRAS G12C ~2x10^4 (Hallin 2020); adagrasib ~5x10^3
10^2 - 10^3 Moderate; clinical possible with high target dwell time Ibrutinib BTK ~5x10^3 (Pan 2007)
< 10^2 Weak; needs warhead optimization Reversibility likely dominates
<= 10 Probably not covalent (or wrong residue) Background rate vs GSH

Intrinsic Reactivity Assays

Before committing to a warhead, measure intrinsic reactivity (off-target risk):

# Generic GSH stability assay readout - measure half-life of warhead with 10 mM GSH
# kinact_GSH from time-course of warhead disappearance
Warhead GSH t1/2 at 10 mM Risk
Chloroacetamide minutes High (reacts with off-target Cys)
Acrylamide hours Moderate
Substituted acrylamide (alpha-Me) days Low (drug-like)
Nitrile days Low
Sulfonyl fluoride hours-days Variable

The "GSH-stable" warhead (t1/2 > 4 hours) is the modern target for druglike TCIs.

Covalent Docking Tools

Tool Approach Strength Fails when
DOCKovalent (London et al 2014 Nat Chem Biol 10:1066) Constraint-based DOCK Free, well-validated Browser-based; small library
GOLD covalent (CCDC) GOLD with covalent constraint Commercial; selectivity License cost
AutoDock 4 covalent AD4 with covalent bond Open source Slower than Vina
CovDock (Schrödinger) Glide-based + covalent Commercial best License cost
MOE covalent Triposite Discovery Commercial License cost
HCovDock (Wu Q, Huang S-Y 2023 Briefings Bioinform 24:bbac559) Hierarchical fragment + covalent Open; supports many residues Newer, less validated
ICM-Pro covalent Active site grid + covalent Commercial; metal centers License cost

For open-source covalent docking, HCovDock (2023) is the modern alternative; DOCKovalent is the longstanding standard.

Example: KRAS G12C Inhibitor Design Workflow

Goal: Decorate a co-crystal scaffold with a cysteine-targeting warhead and rank candidates by covalent efficiency.

Approach: Load scaffold SMILES, enumerate acrylamide-bearing analogs, filter by reactivity selectivity, dock under covalent constraint, and rank by kinact/Ki surrogates.

from rdkit import Chem

# Step 1: scaffold from co-crystal (4LRW or AMG510)
scaffold_smi = 'c1ccc(C(=O)NC2=Nc3c(...)cnc23)cc1'
scaffold = Chem.MolFromSmiles(scaffold_smi)

# Step 2: enumerate analogs with acrylamide warhead
def add_acrylamide(scaffold, attachment_atom_idx):
    """Append acrylamide (-NC(=O)C=C) at a hydrogen position"""
    warhead = Chem.MolFromSmiles('NC(=O)C=C')
    # ... combine via Chem.RWMol or fragment combination
    pass

# Step 3: filter for reactive group selectivity
# Step 4: dock with DOCKovalent / GOLD covalent / HCovDock
# Step 5: rank by kinact/Ki surrogate (compute reactive Michael acceptor reactivity)

Reactivity Surrogates (computed without experiment)

For ranking warheads without wet-lab data:

Descriptor Use case
LUMO energy (DFT) Michael acceptor reactivity (lower LUMO = more reactive)
Electrophile partial charge SN2 reactivity
RDKit rdMolDescriptors.CalcLabuteASA Steric accessibility
AlphaFold3 / Boltz-2 binding pose Geometric fit to reactive Cys

Goal: Approximate Michael-acceptor reactivity from 2D structure without running DFT.

Approach: Parse the SMILES, locate the acrylamide substructure, count alpha-carbon substituents (more substitution lowers LUMO and slows reactivity), and return a negative count as a relative reactivity proxy.

def acceptor_lumo_surrogate(smi):
    # Crude: count alpha-substituents to acrylamide; more substituents = lower reactivity
    mol = Chem.MolFromSmiles(smi)
    if mol is None:
        return None
    acryl_pat = Chem.MolFromSmarts('[CX3](=O)[CX3]=[CX3]')
    matches = mol.GetSubstructMatches(acryl_pat)
    if not matches:
        return None
    # Count substituents on alpha-C
    alpha_c = mol.GetAtomWithIdx(matches[0][1])
    n_subs = len([n for n in alpha_c.GetNeighbors() if n.GetIdx() not in matches[0]])
    return -n_subs  # crude proxy (lower = more reactive)

For real reactivity prediction, DFT calculations (LUMO energy, HOMO-LUMO gap) are needed.

Per-Tool Failure Modes

Wrong warhead for residue

Trigger: Acrylamide warhead targeted at Tyr.

Mechanism: Acrylamides are Cys-selective; do not form bonds with Tyr at physiological pH.

Symptom: No covalent adduct observed despite docking pose.

Fix: Match warhead to residue: acrylamide/chloroacetamide for Cys; sulfonyl fluoride for Lys/Tyr/Ser.

Excessive reactivity (off-target)

Trigger: Chloroacetamide in drug-candidate context.

Mechanism: Too reactive; forms adducts with off-target Cys (e.g., GSH t1/2 < 30 min).

Symptom: Toxicity in cell-based assays; non-specific binding signal.

Fix: Replace with acrylamide (more selective); add alpha-substitution to acrylamide for tunable reactivity.

Geometric mismatch

Trigger: Warhead positioned but Cys-Cβ distance > 6 Å.

Mechanism: Even with reactive warhead, geometric reach matters; Cys side chain has limited reach.

Symptom: No covalent labeling in mass spec despite predicted docking.

Fix: Validate by measuring distance from warhead to Cys-Cβ; redock with constrained covalent bond.

Reversibility unintended

Trigger: Designed irreversible TCI but warhead is reversible.

Mechanism: Nitrile, aldehyde, boronate are reversible; equilibrium with non-covalent.

Symptom: Activity wanes after substrate washout in cellular assays.

Fix: Use truly irreversible warhead (acrylamide, chloroacetamide); or design for reversible covalent intentionally.

kinact/Ki conflation

Trigger: Optimizing for IC50 instead of kinact/Ki.

Mechanism: Compounds with same IC50 differ in covalent efficiency.

Symptom: Apparently identical compounds have different in vivo PK.

Fix: Always measure kinact/Ki (kinetic assay); rank by covalent efficiency.

DOCKovalent over-prediction

Trigger: Default DOCKovalent run.

Mechanism: Covalent constraint forces docking; many ligands "succeed" but are unrealistic.

Symptom: Many compounds pass docking; few label in vitro.

Fix: Post-filter by reactivity (chemoinformatics), geometric fit (Cys-Cβ distance), and PoseBusters.

Reconciliation: Irreversible vs Reversible Covalent

Aspect Irreversible Reversible covalent
Examples KRAS G12C (acrylamide), BTK (ibrutinib) Boronate (bortezomib), aldehyde (calpain inhibitors)
Toxicity profile Off-target Cys labeling potential Off-target equilibrium
Resistance mechanism Mutation of reactive Cys Mutation reduces affinity
Patent / IP Stronger (specific bond) Standard
When to choose If Cys is hot-spot, conserved, druggable If reversibility critical (e.g., proteasome)

Common Errors

Symptom Cause Fix
Warhead not matching SMARTS Different stereochemistry or charged Use canonicalized + neutral SMARTS
DOCKovalent rejects ligand No suitable Cys in pocket Re-check residue accessibility
GSH adduct dominates Warhead too reactive Use less reactive warhead; or alpha-substitute
Off-target labeling in cells Promiscuous warhead Iterate warhead reactivity vs selectivity
Docking pose but no labeling Geometric mismatch Distance check; rotamer search
Reversible inhibitor not acting irreversibly Wrong warhead class Re-check reaction mechanism
HCovDock fails on PROTAC Tool optimized for monomer covalent Use specialized tools for bivalent

References

  • Lonsdale & Ward, Chem. Soc. Rev. 47:3816 (2018) -- covalent revolution review.
  • Singh et al., Nat. Rev. Drug Discov. 10:307 (2011) -- TCI design principles.
  • London et al., Nat Chem Biol 10:1066-1072 (2014) -- DOCKovalent.
  • Wu Q, Huang S-Y et al., Briefings Bioinform. 24:bbac559 (2023) -- HCovDock.
  • Cai et al., J. Cheminformatics 14:39 (2022) -- GOLD covalent toolkit.
  • Backus et al., Nat. Chem. 8:530 (2016) -- proteome-wide covalent ABPP.
  • Pettinger et al., Angew. Chem. Int. Ed. 56:15200 (2017) -- reactive warhead reactivity quantification.
  • Schwartz et al., Nat. Chem. Biol. 10:1006 (2014) -- KRAS G12C disulfide-tethered fragments.

Related Skills

  • chemoinformatics/molecular-io - Parse warhead SMILES
  • chemoinformatics/substructure-search - Warhead SMARTS detection
  • chemoinformatics/virtual-screening - Pre-dock candidate non-covalent fit
  • chemoinformatics/pose-validation - Validate covalent docking
  • chemoinformatics/conformer-generation - Warhead conformer ensembles
  • chemoinformatics/admet-prediction - ADMET of covalent leads
  • chemoinformatics/molecular-descriptors - Reactivity surrogate descriptors