De Novo Peptide Sequencing: How It Works and When to Use It

De novo peptide sequencing reads a peptide's amino acid sequence directly from its tandem mass spectrum, without matching it against a sequence database. You use it when the database has no answer: an organism with no sequenced genome, an antibody with an unknown sequence, or spectra that a standard search leaves unexplained. This guide covers how fragmentation produces readable ladders of ions, how the sequence is inferred and checked, which software does the work, and where the method still falls short.

Bioforgeuk

Source Peptides for Research

BioForgeUK supplies high-purity research peptides for laboratory and scientific research, with products intended for research use only.

Explore research peptides

TL;DR

De novo sequencing derives peptide sequences from tandem mass spectra without a sequence database, by reading mass differences between fragment ions.
It earns its place for unsequenced genomes, therapeutic antibodies, venoms, and spectra that database searches fail to match.
Software proposes sequences, but manual validation is still needed for PTMs, splice variants, SNPs and isobaric residues such as leucine and isoleucine.
Current tools lack the automated false discovery rate control that database searching enjoys, so treat every result as a hypothesis until checked.
The quality of the input peptide, its purity and identity, sets the ceiling for what the spectrum can tell you.

What de novo peptide sequencing actually is
How tandem mass spectrometry breaks a peptide into readable pieces
From spectrum to sequence: how the reading works
Software and computational approaches
Where de novo sequencing is the right choice
A practical workflow
Limits you should plan around
Our perspective: the sequence is only as good as the peptide behind it
Sourcing reference peptides for confirmation work
FAQ
Sources

What de novo peptide sequencing actually is

In a standard proteomics experiment, you digest proteins, run the peptides through a mass spectrometer, and compare each spectrum with theoretical spectra generated from a protein database. The match tells you what the peptide is, but only if the sequence is already in the database.

De novo sequencing drops that dependence. The computational process deduces a peptide's sequence directly from its MS/MS fragment ion spectrum. Instead of asking "which database entry fits this spectrum?", it asks "which sequence of residues would produce this pattern of peaks?". The answer comes from the mass differences between consecutive fragment ions, each of which corresponds to an amino acid residue.

MS/MS peaks mapped to a peptide residue sequence

The practical difference is the search space. A database search chooses among known candidates. De novo sequencing chooses among every possible combination of residues consistent with the precursor mass, which is why it is harder, slower and less certain, and why it is reserved for questions a database cannot answer.

How tandem mass spectrometry breaks a peptide into readable pieces

Everything starts with fragmentation. In the first stage the mass spectrometer isolates a single peptide ion, the precursor. In the second it adds energy so the peptide breaks, and the instrument records the masses of the fragments.

The energy is imparted to the precursor ion to induce covalent bond cleavage, typically along the backbone. Which bond breaks depends on the activation method:

Method	Main bond cleaved	Main ion types
Collision-induced dissociation (CID)	Amide (peptide) bond	b and y ions
Electron-transfer dissociation (ETD)	N-Cα bond	c and z ions
Electron-capture dissociation (ECD)	N-Cα bond	c and z ions

CID primarily breaks amide bonds to give b and y series, while electron-based methods such as ECD and ETD cleave N-Cα bonds to produce c and z ions. The lettering follows the site of breakage: N-terminal fragments are labelled a, b or c, and C-terminal fragments x, y or z.

Why does this matter for de novo work? Because the complementary ion series give you two independent readings of the same sequence. If a b series reads one way and a y series reads the reverse, you can cross-check them. Using more than one activation method, in parallel or in hybrid strategies, often gives a more complete picture of a peptide, and that is especially valuable when you have no database to fall back on.

Complementary b and y ion series cross-check a peptide sequence

From spectrum to sequence: how the reading works

Imagine a clean y-ion series. The first peak sits at a certain mass-to-charge value. The next peak is higher by the mass of one residue, say 57.02 for glycine. The next differs by 71.04, alanine. Walking along the ladder, each gap names one amino acid. Join the names together and you have a sequence tag, or with luck, the whole peptide.

The interpretation is built on series of fragment ions whose mass differences match amino acid residue masses. Software automates this by scoring candidate sequences against the observed peaks. Real spectra are rarely clean, though. Typical complications include:

Missing fragments. If a cleavage site gives no peak, the gap between two observed ions is the sum of two residues, and the software must work out which pairs fit.
Noise and interfering peaks. Chemical noise, co-isolated peptides and neutral losses all add peaks that do not belong to the ladder.
Isobaric residues. Leucine and isoleucine have the same mass. Telling them apart needs characteristic satellite ions such as d, v and w ions, which an experienced analyst looks for by hand.
Modifications. A mass shift on one residue can look like a different amino acid, or two.

Imperfect spectra often yield only partially correct sequence tags, a stretch of confident residues flanked by uncertain mass gaps. A short tag is still useful: it can seed a homology search against related proteins, which is how de novo results often become identifications.

Manual validation still matters

Automation has not removed the human from the loop. Experts check for characteristic peaks and satellite ions to confirm a sequence or separate isomeric residues. The Hunt Lab's practical guide notes that manual validation is needed for proteins with undetermined sequences, PTMs, splice variants or SNPs. Medzihradszky and Chalkley make the same point from another angle: mastering manual interpretation lets researchers evaluate data critically and catch misidentifications that automated engines overlook.

A sensible habit: annotate the spectrum, mark every b, y (or c, z) peak your proposed sequence explains, and look at what is left. A large unexplained fraction of intense peaks usually means the sequence is wrong or incomplete.

Software and computational approaches

De novo software has changed a lot over the past decade, and the field now spans several generations of approach.

Graph-based methods. Tools such as pNovo and Open-pNovo build a spectrum graph, use dynamic programming to find high-scoring paths, and rank candidates with a trained scoring model. Open-pNovo uses a spectrum-graph approach, dynamic programming and RankBoost scoring, and the paper compares it with PEAKS, pNovo+ and Novor. It also supports unexpected modifications, reporting 14 to 250% more modified peptides than traditional tools.

Deep learning models. Transformer-based tools treat the spectrum as an input sentence and the peptide as the output. Casanovo is an open-source example, and its developers describe it as suited to immunopeptidomics, metaproteomics, paleoproteomics and venomics, settings where the peptides of interest may not be in your protein database.

Language-model methods. Newer work uses pre-trained protein language models. PLMNovo (2025) aligns spectra with learned peptide embeddings to improve sequencing accuracy.

The broader trend is clear. De novo methods were long considered niche, but improved algorithms and instruments have widened their use. Still, no model removes the need to examine the evidence behind a call.

Where de novo sequencing is the right choice

De novo sequencing is not a replacement for database searching. For well-annotated organisms and routine identification, a database search is faster and comes with statistical confidence. Reach for de novo when:

There is no reference genome. Work on unsequenced organisms is the classic case. De novo methods can characterise venoms, unsequenced organisms and therapeutic antibodies even when the genome is unavailable.
You need to check database results. It is useful for validating database-search results and investigating unmatched MS/MS spectra.
You suspect variants. Splice variants, SNPs and novel proteoforms are things database searches inevitably miss, according to the Hunt Lab, who still regard de novo sequencing as an essential skill.
You are sequencing an antibody or other unknown protein. The usual strategy is to digest the protein with several enzymes to produce overlapping peptides, then assemble them into contigs. Overlaps resolve ambiguities that a single digest cannot.

A practical workflow

A typical bottom-up de novo experiment follows these steps:

Prepare the sample. Reduce, alkylate and digest. Use more than one protease if you want overlapping coverage.
Clean up. Remove salts and detergents that suppress ionisation.
Acquire MS/MS data. Choose fragmentation to fit the question: CID or HCD for b/y ladders, ETD or ECD for c/z ions and labile modifications, or a combination.
Run de novo software. Keep the top several candidates per spectrum, not just the best score.
Validate. Annotate the spectrum, check complementary series, and resolve isobaric residues manually.
Assemble and cross-check. Align overlapping peptides into contigs, and where possible compare against homologous proteins.
Confirm orthogonally. For anything that matters, confirm with a synthetic peptide: if its spectrum and retention time match the original, the sequence is right.

That last step is where peptide quality becomes a practical concern, and it is the reason we talk about de novo sequencing at all.

Limits you should plan around

Be straightforward with yourself about what the method cannot do yet.

No built-in error control. A major limitation of current software is the lack of automated statistical validation comparable to FDR control in database searches. Your confidence has to come from manual checking and orthogonal evidence.
Complex proteins stay hard. For full-length monoclonal antibodies, a fully confident, absolute sequence remains challenging because of incomplete digestion and post-translational modifications. Current workflows can produce useful contigs, but the authors argue that sample complexity and the need for better PTM integration still stand in the way of complete reconstruction.
Partial results are normal. Expect tags and fragments, not tidy full-length sequences, from imperfect data.

Species-level claims, novel variant claims and anything headed for publication deserve extra scepticism for these reasons.

Our perspective: the sequence is only as good as the peptide behind it

We supply research peptides, so we see de novo sequencing from the other side of the bench: as the people whose products often end up in the instrument. Three things come up repeatedly in our conversations with laboratories.

First, synthetic peptides are the best confirmation tool de novo work has. When software proposes a sequence, making that sequence and comparing spectra and retention behaviour is the cleanest test. That only works if the reference material is what its label says. A peptide with truncation or deletion impurities produces a messy comparison spectrum, and you can lose days chasing a mismatch that is really a purity problem. This is why we work to 99%+ HPLC purity and why we encourage reading the paperwork rather than trusting a headline figure; our guide to reading a peptide Certificate of Analysis explains what to look for, and the HPLC versus UPLC comparison covers how purity is measured.

Second, handling affects the spectrum. Oxidised methionine, deamidated asparagine and aggregated material all show up as extra masses, and a de novo algorithm can mistake them for sequence features. Good reconstitution and storage practice keeps your reference clean; see our peptide handling guidelines for UK research labs.

Third, treat software output as a lead, not a finding. The field's own experts say so, and in our experience the labs with the most reliable results are those that validate by hand and confirm with a standard.

— Neil

Sourcing reference peptides for confirmation work

If your de novo result needs a synthetic reference, BioForgeUK supplies research peptides at 99%+ HPLC purity with Certificates of Analysis, for laboratory use only. Products are not for human or veterinary consumption, and you can read what that label means on our Research Use Only page.

Browse the range from the BioForgeUK home page, and check the CoA for the batch you receive before you run your comparison.

FAQ

Is de novo sequencing better than a database search?

Neither is better in general. A database search is faster and provides statistical error control, so it suits well-annotated organisms. De novo sequencing is the tool when the sequence is not in any database, or when you want to test what a database search may have missed.

Can software tell leucine from isoleucine?

Not from the mass alone, because they are isomers. Analysts look for characteristic satellite ions such as d, v and w, which help differentiate isomeric residues. Without those peaks, the call is usually left ambiguous or resolved by homology.

Which fragmentation method should I use?

CID gives b and y ions and suits most tryptic peptides. ETD or ECD gives c and z ions and is helpful for larger peptides and labile modifications. Because combining activation methods often gives a fuller picture, many labs acquire both for difficult samples.

How do I confirm a de novo sequence?

Synthesise or source the peptide and compare its MS/MS spectrum and chromatographic retention with the original. Matching both is strong evidence. Use a reference of verified purity, and check its CoA before comparing.

Does deep learning make manual checking unnecessary?

No. Models such as Casanovo are powerful, but current de novo tools still lack automated FDR-style validation, so unusual or important calls need inspection of the annotated spectrum.