Peptide Sequencing Via Protein Language Models

TL;DR

The paper addresses the challenge of complete peptide sequencing from partial, experimentally identifiable amino acids. It proposes a transformer-based protein language model, derived from ProtBERT, that is fine-tuned to predict masked residues, simulating Edman/degradation-like constraints on UniRef100 data from three Escherichia species. The approach achieves high per-residue and per-sequence accuracies even with very limited known amino acids and demonstrates biological relevance through AlphaFold-based structural validation, along with insights into cross-species generalizability. This work suggests a probabilistic reconstruction pathway that could accelerate proteomics and structural biology by enabling sequence completion under constrained experimental conditions, with clear directions for wet-lab integration and proteoform-focused diagnostics.

Abstract

We introduce a protein language model for determining the complete sequence of a peptide based on measurement of a limited set of amino acids. To date, protein sequencing relies on mass spectrometry, with some novel edman degregation based platforms able to sequence non-native peptides. Current protein sequencing techniques face limitations in accurately identifying all amino acids, hindering comprehensive proteome analysis. Our method simulates partial sequencing data by selectively masking amino acids that are experimentally difficult to identify in protein sequences from the UniRef database. This targeted masking mimics real-world sequencing limitations. We then modify and finetune a ProtBert derived transformer-based model, for a new downstream task predicting these masked residues, providing an approximation of the complete sequence. Evaluating on three bacterial Escherichia species, we achieve per-amino-acid accuracy up to 90.5% when only four amino acids ([KCYM]) are known. Structural assessment using AlphaFold and TM-score validates the biological relevance of our predictions. The model also demonstrates potential for evolutionary analysis through cross-species performance. This integration of simulated experimental constraints with computational predictions offers a promising avenue for enhancing protein sequence analysis, potentially accelerating advancements in proteomics and structural biology by providing a probabilistic reconstruction of the complete protein sequence from limited experimental data.

Figures (9)

• Figure 1: Frequency distribution across different amino acids and sets in E. coli (top), E. albertii (middle), and E. fergusonii (bottom).
• Figure 1: The 3-fold unmasking inference results with density histograms from (a) E. coli, (b) E. albertii, and (c) E. fergusonii with 4 known amino acids KCYM.
• Figure 2: Overview of pipeline.
• Figure 2: The 3-fold unmasking inference results with density histograms from (a) E. coli, (b) E. albertii, and (c) E. fergusonii with 9 known amino acids KCYMRHWST.
• Figure 3: An illustrated example of structure based model validations.