# How many proteins are there? > Humans have about 20,000 proteins, one per protein-coding gene. UniProt lists 150 million protein records across all life. See what each count measures. Humans have about 20,000 different proteins when each protein-coding gene is counted once. GENCODE release 50, the reference annotation of the human genome, lists **19,442** [protein-coding genes](/guides/how-many-genes-do-humans-have), and UniProt holds 20,431 manually reviewed human protein entries. Across all known life, the UniProtKB database contains about **150 million** protein sequence records, and its archive, UniParc, has stored 1.2 billion distinct sequences. Neither figure is the number of proteins that exist. One gene can give rise to several amino acid sequences, each sequence can carry different chemical modifications, and a single human cell holds billions of protein molecules. That is why published estimates of the human proteome range from 10,000 proteins to several billion. A number quoted without its level, organism and date is not yet an answer, so Table 1 gives the best current figure at each level. Table 1. How many proteins? The answer at each level. The rows count different biological units, so they cannot be added together. Sources: GENCODE v50, UniProt release 2026_03, Aebersold et al. (2018), Dolgalev et al. (2023), Berkeley Lab (2005). What is counted Best current figure Human protein-coding genes 19,442 in GENCODE v50 Reviewed human protein entries 20,431 in UniProtKB/Swiss-Prot Distinct annotated human protein sequences 172,117 in GENCODE v50 Human proteoforms Unknown; about one million per cell type discussed as a mapping target Protein molecules in one human cell About 3.36 billion in a HeLa cell Protein sequence records, all organisms About 150 million in UniProtKB Distinct protein sequences ever archived About 1.2 billion in UniParc Protein types across all life Unknown; the quoted 50 billion is not reproducible ## How many different proteins do humans have? The human count starts with genes. GENCODE release 50 counts 19,442 protein-coding genes on the main chromosomes and leaves out 665 readthrough genes, whose transcripts run across two neighboring genes. Ensembl 116 uses the same gene set but reports 19,878 coding genes, mainly because it also counts the 412 antibody and T-cell receptor gene segments that GENCODE lists separately. UniProt approaches the same set from the protein side. Its 20,431 reviewed human entries roughly follow one entry per gene, with the alternative versions of a protein described inside that entry rather than as separate records. Most of the set has now been seen as protein: the Human Proteome Project reports protein-level evidence for 93.6% of its 19,435-protein [reference human proteome](/guides/human-proteome-size). Sorted by the job they do rather than counted one by one, these proteins fall into [about seven functional types](/guides/types-of-proteins-in-human-body), such as enzymes, transporters and structural proteins. Counted more finely, they give much larger numbers, because a gene can be read in more than one way. A gene's RNA copies, or transcripts, can be spliced differently, joining different segments of the gene, and so encode different amino acid sequences. GENCODE annotates 172,117 distinct protein sequences, known as translations, across its protein-coding genes. Some transcripts encode the same sequence, so the [number of transcripts](/guides/human-transcriptome-size), 278,455 protein-coding models, is larger again. ![Human annotation counts: 19,442 protein-coding genes, 172,117 distinct translations, and 278,455 protein-coding transcripts.](/images/charts/human-proteome-counting-levels.webp '**Figure 1. Three levels of human protein annotation.** The categories overlap and describe different biological units, so they cannot be added together. Source: [GENCODE v50](https://www.gencodegenes.org/human/stats.html).') Table 2. Human genes, translations and transcripts in GENCODE v50. Distinct translations are counted within each gene and then summed, with specific rules for partial coding sequences, so the total is not a global count of unique sequence strings. Sources: GENCODE v50 release statistics and their explanation. Annotation level GENCODE v50 count What is counted Protein-coding genes 19,442 Gene loci under the release's headline counting rules Distinct translations 172,117 Different annotated amino acid sequences Protein-coding transcripts 278,455 RNA transcript models, including full and partial coding sequences All three are annotation counts. They record what the genome is predicted to encode, not experimental proof that each sequence is made as a stable protein. Definitions also change between releases. The roughly 70,000 protein sequences cited in a 2018 review by Aebersold and colleagues came from an older Ensembl annotation, and the gap between that figure and today's reflects differences in annotation coverage and counting rules rather than newly discovered genes. ## How many annotated protein variants have been observed? Protein-level evidence for a gene does not show which of its annotated sequences are made, and no experimental count yet corresponds to the 172,117 translations. The way proteins are measured is a major reason. Most large proteomics experiments cut proteins into short fragments called peptides and identify them by mass spectrometry, which measures the mass of molecules with high precision. A peptide can show that a particular stretch of sequence was made, such as one created by alternative splicing, but that stretch is often shared by several protein variants that differ elsewhere. To see how often this happens, we reanalysed a published table of peptides linked to splicing events in six human cell lines. Of 46,720 distinct observed peptide sequences, 95.6% matched more than one GENCODE v50 reference sequence, and only 4.4% pointed to a single one. Table 3. How specific observed peptides are to GENCODE v50 sequences. Our reanalysis of the supplementary splice-event table from Sinitcyn and colleagues. Each peptide is counted once across cell lines, enzymes and event associations. Sources: Sinitcyn et al. (2023), GENCODE v50. Matching result for an observed peptide Distinct peptide sequences Share of the 46,720 peptides Matches multiple reference sequences 44,655 95.58% Matches exactly one reference sequence 2,038 4.36% No matching reference sequence 27 0.06% The APP gene, which encodes the amyloid precursor protein, offers a concrete example: all 15 APP-associated peptides in our extraction matched multiple reference sequences. That does not undermine the original evidence for the splicing events themselves. Several shared peptides considered together can sometimes pin down a single variant, although we assessed peptides individually and did not attempt that combined inference. The pattern held under stricter rules. Among 30,268 peptides at least nine amino acids long and supported by at least two spectra in one table entry, 96.4% matched multiple reference sequences. Limiting the competing references to full-length protein-coding models on the main chromosomes still left 94.4% of the original peptides with multiple matches. Our analysis therefore measures how ambiguous individual peptides are, not what fraction of human protein variants has been observed. It used the table's splice-event peptides rather than every peptide in the study, and a reference sequence without a match here may have evidence elsewhere. The methodology section describes the method and links the data. ## How many human protein forms are there? Even a complete list of amino acid sequences would not end the count, because each sequence can exist in several molecular forms. A proteoform is one specific form of a protein, defined by its exact sequence together with any chemical modifications. Inherited sequence differences, alternative splicing, cleavage into shorter pieces and the addition of chemical groups, such as the phosphate added in phosphorylation, can all produce distinct proteoforms from a single gene. How many exist in humans is unknown. This is why gene counts cannot settle the size of the human proteome. Aebersold and colleagues discussed one million proteoforms in a single cell type as a plausible depth for future mapping, not as an established count, and a census across tissues, individuals and conditions would have a broader scope still. Possible combinations of modifications also far exceed those that actually occur. The peptide problem returns here too: finding two modified peptides does not show that both modifications sit on the same intact molecule. ## How many proteins are in a cell? Counting physical molecules gives by far the largest numbers, because every copy counts. Dolgalev and colleagues integrated published copy-number estimates for 12,653 distinct proteins in HeLa cells, a widely used human cancer cell line, and arrived at about **3.36 billion** protein molecules per cell. HeLa cells are not a reference for every human cell, but the estimate conveys the scale: on average, each distinct protein is present in roughly a quarter of a million copies, although individual proteins differ widely in abundance. ![Estimated protein molecules per cell: 5.85 million in E. coli, 81.63 million in budding yeast, and 3.36 billion in HeLa.](/images/charts/protein-molecules-per-cell.webp '**Figure 2. Protein copies per cell in three model systems.** The logarithmic axis spans the large differences between the published integrated estimates. Source: [Dolgalev et al. (2023)](https://pmc.ncbi.nlm.nih.gov/articles/PMC9916689/).') Table 4. Protein identities and copies per cell in three model systems. Totals are shown as published in the integrated proteome maps; they are estimates, not counts precise to the molecule. Source: Dolgalev et al. (2023). Cell system Canonical proteins in the integrated map Estimated molecules per cell Escherichia coli 3,852 5,852,319 Budding yeast, Saccharomyces cerevisiae 4,680 81,627,580 HeLa 12,653 3,360,824,528 The same approach puts about 5.9 million protein molecules in an *E. coli* bacterium and 82 million in a budding yeast cell. To build each estimate, the authors selected seven studies per organism, scaled their measurements to literature values for total protein mass per cell and averaged the results. An earlier synthesis of 21 yeast datasets by Ho and colleagues arrived at about 42 million molecules, half the newer figure. Totals like these depend on calibration, growth conditions and which proteins were detected, and they describe an average cell in a population rather than any one cell. Total protein mass also rules out much larger figures. The US National Institute of General Medical Sciences has published an estimate of 10 trillion proteins in each human cell. Ten trillion molecules of even a small 10 kDa protein would weigh about 166 nanograms, more than 600 times the 250 picograms of total protein that Dolgalev and colleagues used for a HeLa cell. A count in the billions fits the protein mass of a human cell; one in the trillions does not. ## How many protein sequences have been catalogued? Beyond humans, the largest curated catalogue of protein sequences is UniProtKB, which in release 2026_03, dated September 2, 2026, held about 150 million records from organisms across the tree of life. Only 575,748 of them, fewer than 0.4%, belong to Swiss-Prot, the manually reviewed section; the rest are in TrEMBL, whose annotation is computational. Any of these records, or a whole reference proteome, can be fetched by accession with [UniProt Download](/app/uniprot-download). Table 5. Protein records in UniProtKB release 2026_03. Checked October 17, 2026. For the API links, counts appear in the X-Total-Results response header and the release in X-UniProt-Release. Sources: Swiss-Prot release statistics and UniProt all-record and unreviewed-record searches. Database section Records in release 2026_03 Annotation status Swiss-Prot 575,748 Manually reviewed TrEMBL 149,430,635 Unreviewed, with computational annotation UniProtKB total 150,006,383 Both sections combined A record count is not a protein count. Reviewed status describes curation, not necessarily direct experimental observation, and UniProt separately grades the evidence that each protein exists, from protein-level evidence through transcript-level evidence to inference from related sequences. An entry can also group several isoforms of a protein, so the number of entries cannot stand in for the number of unique sequences, genes or proteoforms. UniParc casts a wider net. It stores each distinct sequence from the public sequence databases once and never removes an identifier, so its 1,200,368,847 entries in release 2026_03 include sequences that have since been deleted from their source databases. That makes it the closest available count of protein sequences ever recorded, but it covers only organisms and samples that someone has sequenced and submitted. ## How many protein types exist across all life? Nobody knows. The frequently quoted figure of about 50 billion comes from a 2005 Berkeley Lab article that based it on the number of known life forms, without enough detail to reproduce the calculation. A modern estimate would first need a counting rule, since identical sequences shared between species, sequence variants and chemically modified forms could each be grouped or counted separately. The [proteome sizes of individual species](/guides/proteome-sizes-by-species) answer a narrower question with a clearly defined denominator. ## Methodology We checked the database counts on October 17, 2026. UniProt figures come from the UniProtKB and UniParc REST searches linked in the sources, which report release 2026_03; GENCODE v50 and Ensembl 116 figures come from their published release statistics. The 166-nanogram mass in the cell section is our calculation: 1013 molecules × 10,000 g/mol ÷ 6.022 × 1023 molecules/mol ≈ 1.66 × 10−7 g. The 10 kDa mass is deliberately small; a typical human protein is several times heavier, which would widen the gap. For the peptide evidence, we analysed Supplementary Table 3 from Sinitcyn and colleagues' 2023 deep proteome sequencing study, which associates peptides with splicing events in six human cell lines. We extracted distinct peptide sequences with positive spectrum counts and searched them against 241,765 distinct sequences in the full GENCODE v50 translation file. This broad reference set deliberately includes partial models and alternative loci as possible competing matches, so it differs from the headline annotation count. Because mass spectrometry cannot distinguish isoleucine from leucine, we treated the two as identical when matching. We did not reanalyse the spectra or recalculate identification error rates. The [methods and reproduction instructions](/data/guides/protein-isoform-evidence/README.md) document the reference definitions, the sensitivity checks and the 40 spreadsheet cells we excluded because their peptide and spectrum-count lists could not be paired reliably. The [input provenance](/data/guides/protein-isoform-evidence/sources.json), [analysis code](/data/guides/protein-isoform-evidence/analyze.py), [peptide-level results](/data/guides/protein-isoform-evidence/peptides.csv.gz) and [reference-sequence results](/data/guides/protein-isoform-evidence/sequences.csv.gz) are available for inspection and reuse.