Sequence assembly

14 methods · 2016–2026

Sequence assembly: Sequencing a whole protein rather than a peptide. Overlapping peptides are called first, then assembled into full-length sequence, which is how antibodies and protein therapeutics are sequenced without a reference.

Sequencing a whole protein rather than a peptide. Overlapping peptides are called first, then assembled into full-length sequence, which is how antibodies and protein therapeutics are sequenced without a reference.

The earliest of its 14 methods is Protein de novo sequencing by peptide assembly (Wang thesis) (2016); 13 more have followed.

Methods 14
Papers describing them 21
Authors 84
Active 2016-05-13 to 2026-08-13
Deep learning 1 of 14
Kinds post-processor (13), algorithm
Acquisition DDA (13)

Methods (14)

Oldest first, by the paper that describes each one.

  • Protein de novo sequencing by peptide assembly (Wang thesis) (2016): Assembles overlapping de novo sequenced peptides from multiple digests into whole protein sequences such as antibodies.
  • Protein scaffold filling with a reference (2016): Fills missing residues in de novo protein scaffolds against a homologous reference to maximise BLOSUM62 score, with exact and greedy algorithms tested on antibodies.
  • ALPS (2016): Assembles de novo sequenced peptides and their per-residue confidence scores into a de Bruijn graph to reconstruct complete monoclonal antibody heavy and light chains without a template.
  • Top-down protein analysis from de novo strings (2016): Takes a set of de novo amino acid strings derived from top-down tandem mass spectra and analyses the protein sample from them, rather than matching spectra to a database.
  • DiPS (2017): Database-independent Protein Sequencing: de novo sequencing plus assembly that determines full-length protein and antibody sequences without a database.
  • Stitch (2022): Assembles de novo peptides from Casanovo, PEAKS, pNovo and MaxNovo into full antibody sequences, and corrects the two error classes that assembly alone cannot: mass coincidences, where a different residue combination matches the same mass, and I/L ambiguity.
  • PASS (2022): A de novo assembler for short peptide sequences that reconstructs protein targets into contigs of 100 residues or more from LC-MS/MS peptide reads.
  • DBAS (2023): Assembles de novo peptide reads into antibody light and heavy chains using a de Bruijn graph weighted by de novo quality and alignment scores.
  • Fusion (2024): Beam-search assembler that combines overlapping de novo peptides with template information into full antibody sequences, including mixtures of two or three monoclonal antibodies.
  • SequenceAssembler (2025): Post-identification tool that assembles full-length protein sequences by unifying peptide-spectrum matching (PSM) and de novo sequencing outputs from Novor Cloud, PEAKS Studio, and PatternLab for Proteomics; one-click GUI and comparable in performance to Stitch.
  • Generative AI protein scaffold filling (2025): Fills gaps in de novo protein scaffolds with denoising autoencoder, transformer and GPT-2 models, reaching complete accuracy on an antibody scaffold.
  • InstaNexus (2025): End-to-end workflow for reference-free sequencing of full-length protein therapeutics. Multi-protease digestion yields overlapping peptides, InstaNovo sequences them de novo and Winnow rescores, then greedy overlap or de Bruijn graph assembly (default k=7, min overlap 3) reconstructs contigs ranked by a composite score over coverage, N50, scaffold count and identity. Validated on nanobodies, monoclonal antibodies and de novo mini-binders.
  • Homo-Tag-Assembler (2025): Assembles full-length antibody variable regions from de novo peptides using homology-tag fuzzy matching against a probability table of antibody sequences, also separating light and heavy chains and I/L.
  • borgonovo (2026): Reference-free protein sequencer: multi-protease digestion tiles a protein with overlapping peptides, and the redundant de novo reads are assembled into a per-residue consensus by substitution-tolerant alignment and per-column voting. Re-decoding each spectrum under a prior from its consensus position raises amino-acid accuracy. Wraps Casanovo by default but is backend-agnostic.

Papers describing them (21)

Authors (84)

Alberto Santos, Alexandra Gabashvili, Alfred Nilsson, Alon Savidor, Ana Gisele da Costa Neves-Ferreira, Andreas Hougaard Laustsen, Anne Ljungars, Baozhen Shan, Biao Cai, Binhai Zhu (Montana State University), Celso Vitor A. Q. Calomeno, Cheng Ge, Dalia Elinger, Darian Stephan Wolff, Douwe Schulte, Elpida Lytra, Emil Sporre, Erwin M. Schoof, Farong Zhong, Feng Liu, Fredrik Edfors, Hao Wang, Hulyana Brum, Jemma Daniel, Jeroen Van Goey, Jin Lu, Jin Xiao, Jingyi Wang, Joost Snijder, Jordan Sturtz, Kira Vyatkina, Konstantinos Kalogeropoulos, Kushal Badal, Lei Xin, Lennard J. M. Dekker, Letu Qingge, Lin He, Ling Luo, Ljiljana Paša-Tolić, Luis Miguel Muñoz-Gómez, Lukas Käll, M. Ziaur Rahman, Maike Wennekers Nielsen, Marco Reverenna, Marie V. Lukassen, Marlon D. M. Santos, Martijn M. VanDuijn, Michel Batista, Ming Li, Moshit Lindzen, Ngoc Hieu Tran, Nikola Tolić, Ning-Shao Xia, Nuomin Li, Ophir Adiv Tal, Pasquale D. Colaianni, Paulo C. Carvalho, Pavel A. Pevzner, Qing Xu, Quan Yuan, Ren Kong, René L. Warren, Richard Annan, Richard Hemmi Valente, Rodrigo S. C. Brant, Rong Wang, Rongshan Yu, Rotem Barzilay, Shan Chang, Si Wu, Suthimon Thumtecho, Theo M. Luider, Timothy P. Jenkins, Weiwei Peng, Wenbin Jiang, Xiaowen Liu, Xu Wang, Yi Lu, Yishai Levin, Yongqian Zhang, Yosef Yarden, Yubo Song, Yueting Xiong, Zhenjian Jiang

Seen in the charts

Back to the full map

Back to top