Segments of DNA which code for specific proteins are called coding sequences, and in most human genes the protein-specifying portions are carried by exons—the stretches of a gene that survive processing and are spliced together into mature messenger RNA before translation. The mature mRNA is then read three nucleotides at a time, each triplet specifying one amino acid or a stop signal. The full set of protein-coding instructions in a human genome spans roughly 1.2 percent of its 3 billion base pairs.

From Gene to Protein: The Actual Machinery

The central mechanism starts with transcription, not translation. RNA polymerase II reads a gene's DNA template strand and builds a complementary pre-mRNA molecule, copying both exons and the intervening introns. In humans, introns are usually far longer than exons. A typical human gene might span 27,000 base pairs, yet its mature mRNA could be only 1,300 nucleotides long after introns are removed and exons are joined.

Splicing is performed by the spliceosome, a large RNA-protein machine that recognizes short boundary signals at exon-intron junctions. The most conserved of these are the 5′ splice donor site, which usually begins with GT, and the 3′ splice acceptor site, which usually ends with AG. The spliceosome cuts, lassos the intron into a lariat structure, and ligates the flanking exons. This is why exons are sometimes defined as the sequences retained in mature RNA, not simply as the sequences that encode protein.

Translation then converts the exon-built mRNA into a polypeptide. Ribosomes read codons—three-base units—and transfer RNAs deliver the matching amino acid. A codon table maps 64 possible triplets onto 20 amino acids plus three stop codons. Because the reading frame is fixed, deleting or inserting a single nucleotide shifts every downstream codon, which is why frameshift mutations are typically more damaging than substitutions.

The counterintuitive detail is that not every exon is purely protein-coding. Many exons contain untranslated regions at the mRNA's 5′ and 3′ ends, and some exons encode only regulatory elements. The protein-specifying portion is the coding sequence within the mRNA, called the CDS, and it may begin and end partway through the first and last exons.

Alternative splicing lets one gene produce multiple proteins. By including or skipping particular exons, a cell can generate distinct messenger RNAs from the same DNA. Estimates suggest that more than 90 percent of human multi-exon genes undergo some form of alternative splicing, though the functional significance of every isoform is not established.

Quantified Context: How Much DNA Actually Codes

The human genome contains about 3.1 billion base pairs per haploid set. Protein-coding exons occupy roughly 1.2 percent of that total, approximately 35 to 40 million base pairs. The remaining sequence includes introns, regulatory regions, repetitive elements, and noncoding RNA genes.

Protein-coding genes number between 19,000 and 20,000. The average coding sequence is about 1,100 to 1,500 base pairs, encoding a protein of roughly 350 to 500 amino acids. Average gene length is closer to 27,000 base pairs, so introns account for the large majority of most genes' span.

The ENCODE project, reported in 2012, found that about 80 percent of the genome shows some biochemical activity, such as transcription factor binding or chromatin marks. This figure is often misread. Biochemical activity is not the same as protein-coding function; only the small coding fraction directly specifies amino acid sequences.

Mutation rates provide another scale. The human germline substitution rate is roughly 1.2 × 10⁻⁸ per base pair per generation, meaning each newborn carries about 60 to 70 new single-nucleotide variants. Most fall outside coding exons, and of those within coding regions, many are synonymous—they change a codon without changing the amino acid.

Edge Cases: Pseudogenes, Overlapping Genes, and Noncoding RNA

Pseudogenes are DNA segments that resemble protein-coding genes but carry disabling mutations. The human genome contains thousands of processed and unprocessed pseudogenes. They are not translated into functional proteins, so they illustrate why sequence similarity alone does not establish coding status.

Overlapping genes share DNA sequence but use different reading frames or strands. Some viruses, notably hepatitis B, pack coding information into overlapping open reading frames. In humans, overlapping genes are rare but documented. A single base pair can therefore contribute to two different proteins depending on how it is read.

Noncoding RNAs are transcribed but not translated. Ribosomal RNA, transfer RNA, microRNA, and long noncoding RNA all derive from DNA, yet their function is the RNA molecule itself, not a protein product. Calling them "noncoding" distinguishes them from messenger RNA, not from function.

Mitochondrial DNA is a separate exception. The human mitochondrial genome is about 16,569 base pairs and contains 37 genes, 13 of which encode proteins of the oxidative phosphorylation system. Mitochondrial codons differ slightly from nuclear ones; for example, UGA normally signals stop in the nucleus but encodes tryptophan in mitochondria.

Red flags that a claimed "coding" segment may not be what it appears:

Sequence similarity to a known gene without an intact open reading frame. A predicted coding region lacking conserved splice donor and acceptor signals. Transcripts detected by RNA sequencing but never associated with ribosome footprints. Variants annotated as "coding" that fall in untranslated exonic regions. Databases conflating gene span with coding sequence length.

Misconceptions That Persist

The largest misconception is that a gene equals a protein. A gene is a DNA locus that produces one or more functional RNA products; only some of those products are translated. The gene's span includes introns and regulatory regions that never reach the ribosome.

A second error is equating "exon" with "coding sequence." Exons are retained in mature RNA, and untranslated exons exist at both ends of most messenger RNAs. Conversely, the coding sequence can begin after the first exon and end before the last.

A third misconception is that noncoding DNA is "junk." Regulatory elements, enhancers, promoters, and noncoding RNAs are functionally important, even though they do not encode amino acid sequences. The old junk-DNA framing overstated the case in both directions.

A fourth error concerns the 98.5 percent figure often quoted for noncoding DNA. That number refers to the fraction not occupied by protein-coding exons. It does not mean 98.5 percent of the genome has no function, and it does not mean all of it is regulatory.

A fifth misconception is that a single mutation in an exon always changes protein function. Synonymous variants, variants in untranslated exons, and changes that preserve amino acid chemistry can be benign. Clinical interpretation requires more than the label "coding."

Below is a comparison of the major functional classes of DNA sequence.

Sequence class Retained in mature mRNA? Encodes amino acids? Approximate share of human genome Typical role
Protein-coding exon Yes Yes, in part About 1.2 percent Specifies polypeptide sequence
Untranslated exon Yes No Small fraction of exonic sequence Regulation, mRNA stability, translation control
Intron No, usually removed No Roughly a quarter of gene span Splicing signals, regulatory elements
Promoter and enhancer Not transcribed into mRNA No Dispersed, hard to total Controls transcription timing and level
Noncoding RNA gene Produces its own RNA No Several percent Structural, catalytic, regulatory RNA
Pseudogene Variable No functional product Thousands of loci Evolutionary relic, occasional regulatory role
Mitochondrial coding gene Yes, within organelle Yes 13 protein genes Oxidative phosphorylation subunits

Practical Actions for Reading Genetic Claims

When you encounter a statement that a segment "codes for a protein," verify what level of evidence supports it. A predicted open reading frame is a hypothesis; ribosome profiling, proteomic detection, and conservation across species are stronger evidence.

Check the units and coordinates. Gene span, transcript length, and coding sequence length are three different numbers, and press releases routinely mix them. A gene listed at 100,000 base pairs may yield a 2,000-nucleotide messenger RNA with a 1,500-base-pair coding region.

Look for the transcript accession and the annotated CDS boundaries. Public databases such as RefSeq and Ensembl label coding regions explicitly. If the coding start and stop are not defined, the "coding" label is provisional.

For clinical or research interpretation of a variant, confirm the transcript used for annotation. The same genomic variant can be coding in one isoform and intronic in another. Mismatched transcript references are a common source of false pathogenicity calls.

Concrete thresholds that help frame interpretation:

A coding sequence is generally considered to require an open reading frame of at least 100 codons to be plausibly protein-coding, though shorter functional peptides exist. Conservation scores above roughly 0.5 on common phyloP scales suggest purifying selection, consistent with function. A variant allele frequency below 1 percent in population databases suggests rarity, not automatically pathogenicity. Ribosome footprint density above background supports translation, but absence of footprints does not prove noncoding status.

Reconciling thresholds matters because no single cutoff decides function. Open reading frame length, conservation, ribosome occupancy, and proteomic evidence each answer a different question. Treat them as converging lines of evidence rather than gatekeepers.

Limitations and What Remains Uncertain

Annotation of the human genome is still incomplete. The 1.2 percent coding fraction is a well-supported estimate, but the number of functional noncoding elements remains debated. Different projects use different definitions of function, which produces different percentages.

Alternative splicing complicates any simple count of "protein-coding segments." A single exon may be coding in one isoform and skipped in another. Consequently, the coding sequence is a property of a transcript, not of the DNA segment in isolation.

Prediction tools can misclassify short open reading frames. Some peptides under 100 amino acids are functional, and some predicted coding regions have no detectable protein product. Databases update annotations as evidence accumulates, so any snapshot can be outdated.

Variant interpretation guidelines from professional genetics bodies emphasize that "coding" is only one input. Functional evidence, segregation data, and population frequency all contribute. No single label substitutes for a structured assessment.

For questions about your own genetic test results, variant interpretation, or inherited condition risk, this article is general educational information, not a diagnosis or treatment plan. A clinical geneticist or genetic counselor can interpret results in the context of your history. If you face an urgent medical concern, contact local emergency services.