# How many covalent drugs are approved?

> A covalent drug forms a chemical bond with its target. ProteinIQ checked 2,104 approved small-molecule drugs in ChEMBL 37 and found 173 that act this way, 8.2% of the set, and 36 (1.7%) that were designed as targeted covalent drugs. Published counts of 50 to 128 differ mainly in whether beta-lactams and DNA alkylators are included.

Aspirin and ibrutinib have little in common except one step in their mechanism. Aspirin hands an acetyl group to a serine inside cyclooxygenase. Ibrutinib bonds to a cysteine in Bruton's tyrosine kinase. In both cases the drug is no longer a guest in the protein; it has become part of it. Penicillin, omeprazole and the first drug against mutant KRAS, sotorasib, work the same way.

How many approved drugs are covalent depends on what you count. We checked all 2,104 approved small-molecule drugs in ChEMBL 37 and found **173** that bond to their therapeutic target, 8.2% of the set. Only **36** of them, 1.7%, are targeted covalent drugs, built around a reactive group that was chosen to hit one residue of one protein. Published counts, which run from 50 to 128, differ mainly in whether old drugs such as beta-lactams and DNA alkylators are counted and in which regulator's approvals are included.

## What makes a drug covalent?

Most drugs hold on to their target through hydrogen bonds, charge attraction and snug packing, and they come off again. A covalent drug shares electrons with an atom of the target, and undoing that bond takes a chemical reaction. The drug needs a reactive group, called a warhead, and the protein needs a nucleophile in reach: usually the thiol of a cysteine or the hydroxyl of a serine, and less often a threonine, a lysine or the free end of the chain.

Modern covalent drugs bind in two steps. The molecule first docks reversibly, like any inhibitor, and the warhead reacts only if that docking puts it next to the right residue. Two numbers describe the process: KI, the affinity of the first step, and kinact, the rate of the reaction. Because the effect grows with time, a single IC50 depends on how long the assay ran, and covalent inhibitors are compared by kinact/KI instead; the guide to [IC50 versus Ki](/guides/ic50-vs-ki) covers why assay-dependent values mislead.

The bond may or may not last. Acrylamides add to a cysteine for good, and the protein is restored only when the cell replaces it. Nitriles, aldehydes, ketoamides and boronic acids form bonds that open again, so the drug can leave. Reversible covalent drugs in our set include nirmatrelvir, bortezomib, vildagliptin and voxelotor.

<caption><strong>Table 1. Warheads among the 173 approved covalent drugs.</strong> Counts are drugs in ChEMBL 37 that meet the broad definition below, grouped by the reactive group that bonds to the target. Sources: ProteinIQ analysis of ChEMBL 37 (2026).</caption>
<thead>
<tr><th>Warhead</th><th>Reacts with</th><th>Bond</th><th>Examples</th><th>Drugs</th></tr>
</thead>
<tbody>
<tr><td>Beta-lactam</td><td>Serine</td><td>Irreversible</td><td>Penicillin G, ceftriaxone, meropenem</td><td>72</td></tr>
<tr><td>Acrylamide and butynamide</td><td>Cysteine</td><td>Irreversible</td><td>Ibrutinib, osimertinib, sotorasib</td><td>16</td></tr>
<tr><td>Fumarate and cyanoenone</td><td>Cysteine</td><td>Irreversible or reversible</td><td>Dimethyl fumarate, omaveloxolone</td><td>4</td></tr>
<tr><td>Nitrile</td><td>Cysteine or serine</td><td>Reversible</td><td>Nirmatrelvir, vildagliptin</td><td>4</td></tr>
<tr><td>Ketoamide and aldehyde</td><td>Serine or protein end</td><td>Reversible</td><td>Telaprevir, voxelotor</td><td>3</td></tr>
<tr><td>Boronic acid and oxaborole</td><td>Threonine, serine or tRNA</td><td>Reversible</td><td>Bortezomib, vaborbactam, tavaborole</td><td>4</td></tr>
<tr><td>Diazabicyclooctane</td><td>Serine</td><td>Reversible</td><td>Avibactam, relebactam</td><td>3</td></tr>
<tr><td>Epoxide, epoxyketone and beta-lactone</td><td>Threonine, cysteine or serine</td><td>Irreversible</td><td>Carfilzomib, fosfomycin, orlistat</td><td>3</td></tr>
<tr><td>Aryl carbamate and organophosphate</td><td>Serine</td><td>Slowly reversible or irreversible</td><td>Neostigmine, echothiophate</td><td>9</td></tr>
<tr><td>Sulfinyl benzimidazole, thienopyridine and disulfide</td><td>Cysteine, after activation</td><td>Irreversible</td><td>Omeprazole, clopidogrel, disulfiram</td><td>10</td></tr>
<tr><td>Propargylamine and hydrazine</td><td>Flavin cofactor</td><td>Irreversible</td><td>Selegiline, phenelzine</td><td>9</td></tr>
<tr><td>Fluoropyrimidine and azacytosine</td><td>Cysteine, after incorporation</td><td>Irreversible</td><td>Fluorouracil, decitabine</td><td>7</td></tr>
<tr><td>Mustard, nitrosourea, aziridine and other alkylators</td><td>DNA guanine</td><td>Irreversible</td><td>Melphalan, carmustine, thiotepa</td><td>24</td></tr>
<tr><td>Others, one to three drugs each</td><td>Various</td><td>Irreversible</td><td>Aspirin, vigabatrin, phenoxybenzamine</td><td>5</td></tr>
</tbody>

## Targeted covalent inhibitors and the older covalent drugs

Most covalent drugs were not designed to be covalent. Penicillin G, first approved in 1947 in ChEMBL's records, acylates a serine in the bacterial enzymes that build the cell wall. Aspirin acetylates cyclooxygenase, and the mustard drugs alkylate DNA. Dalton and colleagues, who compiled the most detailed list of FDA-approved covalent drugs, point out that the earliest of them date from the 1940s, when the mechanism may not yet have been known.

Then the field turned away from the class for decades. Drug makers feared that a reactive molecule would bind unrelated proteins and cause toxicity, and Singh and colleagues argued in 2011 that this fear did not fit the historical record of covalent drugs that worked.

The turn came with targeted covalent inhibitors, which attach a mild warhead to a scaffold chosen for its affinity for one protein. Singh counted eight approved in the decade before 2022, among them drugs that changed treatment for lung cancer and chronic lymphocytic leukemia.

## How many covalent drugs are approved?

We counted under two definitions. The broad definition counts a drug if it, or an active metabolite made in the body, forms a covalent bond with its therapeutic target as part of the accepted mechanism, reversible or not. The target can be a protein, DNA or RNA. A drug that only forms an adduct with a freely dissociating coenzyme, such as isoniazid with NAD or finasteride with NADP, is not counted.

The strict definition keeps the broad drugs whose bond to a defined residue of a protein is the designed basis of the activity: a recognition scaffold plus a reactive group that was installed or tuned against that target, confirmed by structure or mass spectrometry. Drugs found by phenotypic screening, by chance or as natural products before the mechanism was understood stay in the broad set only, however well their chemistry is known today.

<caption><strong>Table 2. Approved small-molecule drugs that act covalently.</strong> Population is 2,104 approved small-molecule parent compounds in ChEMBL 37, approved in at least one jurisdiction ChEMBL tracks. Sources: ProteinIQ analysis of ChEMBL 37 (2026).</caption>
<thead>
<tr><th>Group</th><th>Drugs</th><th>Share of 2,104</th></tr>
</thead>
<tbody>
<tr><td>Broad definition</td><td>173</td><td>8.2%</td></tr>
<tr><td>Broad, beta-lactams only</td><td>72</td><td>3.4%</td></tr>
<tr><td>Broad, other protein targets</td><td>76</td><td>3.6%</td></tr>
<tr><td>Broad, DNA alkylators and crosslinkers</td><td>24</td><td>1.1%</td></tr>
<tr><td>Broad, tRNA (tavaborole)</td><td>1</td><td>0.05%</td></tr>
<tr><td>Broad, without beta-lactams</td><td>101</td><td>4.8%</td></tr>
<tr><td>Strict definition</td><td>36</td><td>1.7%</td></tr>
<tr><td>Strict, irreversible bond</td><td>21</td><td>1.0%</td></tr>
<tr><td>Strict, reversible bond</td><td>15</td><td>0.7%</td></tr>
</tbody>

The beta-lactam antibiotics are the largest single block: 68 penicillins, cephalosporins, carbapenems and monobactams, plus four beta-lactamase inhibitors, so 72 of the 173 broad drugs. Leave them out and 101 remain. All 36 strict drugs bond to a protein, and their targets are 19 cysteines, 12 serines, 3 threonines, one lysine and the N-terminal valine of hemoglobin, which voxelotor binds.

Published counts are lower than ours and do not agree with each other. Sutanto and colleagues wrote in 2020 that at least 50 FDA-approved drugs act as covalent inhibitors. De Vita's 2021 perspective tabulated 88, counting beta-lactams, nonspecific DNA alkylators and drugs that act on cofactors. Brown and colleagues put the figure at no fewer than 75 drugs with regulatory approval.

The newest list, by Dalton and colleagues, covers all FDA-approved small-molecule drugs with a covalent mechanism; a 2025 JACS Au paper that uses it reports 128 compounds, about 7% of FDA-approved therapies.

<caption><strong>Table 3. Published counts of approved covalent drugs and the ProteinIQ counts.</strong> The definitions and approval windows differ, so the numbers are not interchangeable. The 128 and the 7% are as reported by the citing paper; the list itself is not open. Sources: Sutanto et al. (2020), De Vita (2021), Lee and Park (2022), Roskoski (2025), Brown et al. (2026), Xue et al. (2025), ProteinIQ (2026).</caption>
<thead>
<tr><th>Source</th><th>Count</th><th>What it counts</th></tr>
</thead>
<tbody>
<tr><td>Sutanto et al., 2020</td><td>At least 50</td><td>FDA-approved drugs that act as covalent inhibitors</td></tr>
<tr><td>De Vita, 2021</td><td>88</td><td>FDA-approved covalent drugs, including beta-lactams and DNA alkylators</td></tr>
<tr><td>Brown et al., 2026</td><td>At least 75</td><td>Covalent drugs with regulatory approval</td></tr>
<tr><td>Dalton et al., 2025</td><td>128 (about 7%)</td><td>FDA-approved small-molecule drugs with a covalent mechanism</td></tr>
<tr><td>Lee and Park, 2022</td><td>19 (4.4%)</td><td>Covalent drugs among FDA approvals from 2012 to 2021</td></tr>
<tr><td>Roskoski, 2025</td><td>11 of 88</td><td>FDA-approved protein kinase inhibitors that form irreversible bonds</td></tr>
<tr><td>ProteinIQ, 2026</td><td>173 broad, 36 strict</td><td>Approved small molecules in ChEMBL 37 under our two definitions</td></tr>
</tbody>

Our broad count sits above Dalton's for reasons we can list but not test. ChEMBL records approvals from regulators other than the FDA, keeps withdrawn drugs, lists five prodrug esters of beta-lactams as separate parents, and we accept drugs that need metabolic activation or act on the flavin of monoamine oxidase. The Dalton list is behind a paywall, so we could not compare drug by drug. The share is close, 8.2% against about 7%.

The 4.4% in the table is a different kind of number. It is the share of recent approvals that were covalent, not the share of all marketed drugs, so it should be compared with a decade window. Over the same years, 2012 to 2021, we find 17 strict and 27 broad covalent drugs among 302 approved small molecules, 5.6% and 8.9%. Lee and Park counted 19 covalent drugs among FDA approvals that also include biologics.

One more figure travels widely. A 2020 review states in its abstract that about 30% of marketed drugs are covalent inhibitors. The number appears to echo an older estimate, that roughly 30% of drugs acting on an enzyme target inhibit it irreversibly, which has a much smaller denominator than all drugs.

## Which warheads and targets

Beta-lactams make up 42% of the broad set, and DNA alkylators and Michael acceptors, the group that holds the acrylamides, follow. The rest is a long tail of families with seven to ten drugs each.

![Horizontal bar chart of the 173 approved covalent drugs by warhead family: beta-lactams 72, DNA alkylators 24, Michael acceptors 20, disulfide-forming drugs 10, cholinesterase inhibitors 9, MAO inhibitors 9, boronic acids and diazabicyclooctanes 7, nucleoside analogs 7, nitriles and ketoamides 6 and other 9](/images/charts/approved-covalent-drugs-by-warhead-family.webp '**Figure 1. Approved covalent drugs by warhead family.** ProteinIQ analysis of ChEMBL 37 under the broad definition. DBOs are diazabicyclooctane beta-lactamase inhibitors such as avibactam. Other holds epoxides, a beta-lactone, an aryl ester, a haloalkylamine, an aldehyde and three enzyme-activated inactivators. Source: [ProteinIQ analysis of ChEMBL 37 (2026)](/data/guides/approved-covalent-drugs/summary.json).')

By target, the strict drugs concentrate where a reactive residue sits in a druggable pocket. Thirteen hit protein kinases and nine hit proteases, a group that includes the proteasome and DPP-4 as well as the viral proteases of hepatitis C and SARS-CoV-2. Seven act on bacterial cell-wall enzymes. The broad set shifts toward the bacterial enzymes and DNA, which reflects the weight of the older antibiotics and cancer drugs. The [drug targets](/guides/drug-targets) guide shows how the same classes dominate approved drugs overall.

<caption><strong>Table 4. Approved covalent drugs by target class.</strong> Target class is assigned by hand from the covalent target. KRAS and other proteins holds KRAS, exportin-1, KEAP1 and hemoglobin. Sources: ProteinIQ analysis of ChEMBL 37 (2026).</caption>
<thead>
<tr><th>Target class</th><th>Strict</th><th>Broad</th><th>Strict examples</th></tr>
</thead>
<tbody>
<tr><td>Protein kinases</td><td>13</td><td>13</td><td>Ibrutinib, osimertinib, ritlecitinib</td></tr>
<tr><td>Proteases</td><td>9</td><td>9</td><td>Bortezomib, nirmatrelvir, vildagliptin</td></tr>
<tr><td>Bacterial cell-wall enzymes</td><td>7</td><td>77</td><td>Avibactam, tazobactam</td></tr>
<tr><td>KRAS and other proteins</td><td>5</td><td>8</td><td>Sotorasib, selinexor, voxelotor</td></tr>
<tr><td>Other enzymes</td><td>2</td><td>32</td><td>Vigabatrin, eflornithine</td></tr>
<tr><td>DNA</td><td>0</td><td>24</td><td>None</td></tr>
<tr><td>Proton pump and other transporters</td><td>0</td><td>6</td><td>None</td></tr>
<tr><td>GPCRs</td><td>0</td><td>4</td><td>None</td></tr>
</tbody>

## When covalent drugs became designed

Before 1980 no approved drug meets the strict definition. The first are the penicillanic acid sulfone sulbactam in 1986 and the ornithine decarboxylase inhibitor eflornithine in 1990, both designed to inactivate one enzyme. Bortezomib followed in 2003, and from 2010 onward strict drugs account for a growing slice of approvals: 6.0% of the drugs first approved in the 2010s, 16 of 268, and 7.2% of those approved from 2020 to 2024, 11 of 152.

The broad share did not rise in the same way. It swings between 6% and 12% by decade, driven by the antibiotic boom of the 1970s and 1980s. Without the beta-lactams it climbs from 2.3% in the 1970s and 1980s to 8.6% in the 2020s. What changed is who makes covalent drugs and why: serendipity and natural products gave way to design.

![Line chart of the share of approved small-molecule drugs that are covalent by decade of first approval under three definitions. The broad share stays between 5% and 12%, the broad share without beta-lactams rises from 2% in the 1980s to 9% in the 2020s, and the strict share rises from 0% before 1980 to 7% in the 2020s](/images/charts/approved-covalent-drugs-share-by-decade.webp '**Figure 2. Covalent drugs among new approvals, by decade of first approval.** ProteinIQ analysis of 1,805 approved small-molecule parents in ChEMBL 37 with a recorded first approval year; the 2020s cover 2020 to 2024. Source: [ProteinIQ analysis of ChEMBL 37 (2026)](/data/guides/approved-covalent-drugs/summary.json).')

## Kinase inhibitors and KRAS

Kinase inhibitors show the change most clearly. By the end of 2012, ChEMBL records 21 approved drugs in ATC group L01E (protein kinase inhibitors), and none of them is covalent. Afatinib and ibrutinib arrived in 2013. Since then 12 of the 57 new kinase inhibitors, 21%, have been covalent, from osimertinib in 2015 to lazertinib in 2024.

![Line chart of cumulative approved protein kinase inhibitors from 2010 to 2024, rising from 12 to 78, and of covalent kinase inhibitors, which start in 2013 and reach 12](/images/charts/approved-covalent-kinase-inhibitors-by-year.webp '**Figure 3. Kinase inhibitors approved to date.** Protein kinase inhibitors are ATC group L01E in ChEMBL 37, which includes the mTOR inhibitors sirolimus and everolimus. The covalent count excludes ritlecitinib, a JAK3 inhibitor outside that group. Source: [ProteinIQ analysis of ChEMBL 37 (2026)](/data/guides/approved-covalent-drugs/summary.json).')

Roskoski counts eleven irreversible covalent drugs among the 88 kinase inhibitors the FDA has approved. Our 13 kinase drugs are his eleven plus olmutinib and tirabrutinib, which were approved in Korea and Japan but not by the FDA. All of them carry an acrylamide or a butynamide, which adds to a cysteine next to the ATP pocket: Cys797 in EGFR and Cys481 in BTK.

KRAS is the other case. It was long regarded as a protein without a pocket, and the first covalent KRAS inhibitors entered clinical trials in 2019. They target the cysteine that the G12C mutation creates, so the warhead reacts only with the cancer protein. Sotorasib was approved in 2021 and adagrasib in 2022, and Brown and colleagues list KRAS, EGFR and BTK as the famous successes of the field.

ChEMBL 37 stops at drugs first approved in 2024, so it misses the newest ones. Four more covalent kinase inhibitors have FDA approval that ChEMBL 37 lacks, and the data file also lists drugs approved only in China and the reversible covalent BTK inhibitor rilzabrutinib. Adding the four takes the strict count of 36 to at least 40.

<caption><strong>Table 5. Covalent kinase inhibitors approved by the FDA after the ChEMBL 37 cut-off.</strong> Not part of the 173 and 36. First approval is the earliest date we found. Sources: FDA (2025, 2026), Novartis (2025).</caption>
<thead>
<tr><th>Drug</th><th>First approval</th><th>Target</th><th>Bond</th></tr>
</thead>
<tbody>
<tr><td>Sunvozertinib</td><td>China 2023, FDA 2025</td><td>EGFR exon 20 insertions</td><td>Irreversible</td></tr>
<tr><td>Zongertinib</td><td>FDA 2025</td><td>HER2 kinase domain mutants</td><td>Irreversible</td></tr>
<tr><td>Remibrutinib</td><td>FDA 2025</td><td>BTK</td><td>Covalent</td></tr>
<tr><td>Lirafugratinib</td><td>FDA 2026</td><td>FGFR2 fusions</td><td>Irreversible</td></tr>
</tbody>

## Using the list in drug discovery

Covalent drugs look different from the average approved drug. Seven of the 36 strict drugs, 19%, break two or more of Lipinski's limits, against 13.4% of all approved small molecules, and their median weight is 454 Da; the [rule of five](/guides/lipinski-rule-of-five) guide explains the limits.

Many warheads also trigger structural alerts. A screening library cleaned with the [Brenk filter](/app/brenk-filter) or the [PAINS filter](/app/pains-filter) may have lost its covalent leads, so a project that wants them should decide which alerts to keep.

ProteinIQ does not host covalent docking. [AutoDock Vina](/app/autodock-vina), [GNINA](/app/gnina) and [DiffDock](/app/diff-dock) place a ligand without forming the bond, so for a covalent ligand they model the first, reversible step.

That is useful for ranking scaffolds against a pocket that contains a cysteine or serine, provided you measure the distance from the warhead to the residue yourself. [PoseBusters](/app/posebusters) checks whether a docked pose is chemically and physically valid before you interpret it. The [molecular docking](/use-cases/molecular-docking) page describes the wider workflow.

## How we counted

We started from the 2,104 approved small-molecule parents of our [Lipinski analysis](/guides/lipinski-rule-of-five), taken from ChEMBL 37 (released May 1, 2026) through its web services on October 29, 2026. ChEMBL gives max_phase 4 to compounds approved in at least one jurisdiction it tracks, so the set is broader than the [FDA's own list](/guides/how-many-fda-approved-drugs). It leaves out biologics, metal-containing drugs such as cisplatin, drugs without a single defined structure and mixtures.

We searched every structure with RDKit 2025.03.5 for 34 warhead patterns, including acrylamides, butynamides, nitriles, boronic acids, aldehydes, epoxides, beta-lactams, beta-lactones, disulfides, nitrogen mustards, nitrosoureas and aryl carbamates. We also searched ChEMBL's mechanism records for words such as covalent, irreversible and alkylating, and added drugs known to act covalently through groups the patterns cannot see, such as the proton pump inhibitors. That gave 228 candidates.

We reviewed each candidate against its label, ChEMBL's mechanism and the literature, and recorded the decision and its basis. Of the 228, 173 meet the broad definition and 55 do not, such as the acrylamide-like histone deacetylase inhibitors belinostat and panobinostat, which chelate zinc, and finasteride, whose adduct is with a coenzyme. For 153 of the 173 the adduct with the target was shown by structure or mass spectrometry or is stated on the label; for the other 20 the mechanism is accepted but the adduct is not directly characterized.

Several calls shape the numbers. Sulbactam, tazobactam and enmetazobactam count as strict because they were designed as beta-lactamase inactivators; without them the strict count is 33. Mobocertinib and voxelotor are counted although both have been withdrawn. Four drugs the Lipinski population excludes, tranylcypromine and three platinum drugs, would meet the broad definition and are listed in the supplementary file. Decades come from ChEMBL's first approval year, which is missing for 14 of the 173.

The [broad-set table (CSV)](/data/guides/approved-covalent-drugs/covalent-drugs.csv), [all 228 candidates with decisions (CSV)](/data/guides/approved-covalent-drugs/covalent-drug-candidates.csv), [hand-review decisions (TSV)](/data/guides/approved-covalent-drugs/covalent-drugs-review.tsv), [drugs outside ChEMBL 37 (TSV)](/data/guides/approved-covalent-drugs/covalent-drugs-outside-chembl.tsv), [summary of every figure (JSON)](/data/guides/approved-covalent-drugs/summary.json), [methods (README)](/data/guides/approved-covalent-drugs/README.md) and [analysis script](/data/guides/approved-covalent-drugs/analyze-covalent-drugs.py) reproduce all of our numbers.

Because ChEMBL is licensed CC BY-SA 3.0, the data files are shared under the same license with credit to ChEMBL and ProteinIQ. We last checked them on October 29, 2026.