Abstract
Elucidating antibody sequences by mass spectrometry (MS)-based de novo sequencing is essential but challenging. Herein, we propose XA-Novo, an accurate, robust, high-throughput de novo sequencing solution that integrates a single-pot multi-enzymatic gradient digestion method with a beam search-based Fusion assembler to accurately decipher complete antibody sequences at both the sample preparation stage and the data analysis phase. We first benchmark XA-Novo using data from multiple known antibodies across various species, demonstrating that it outperforms commercial solutions in identification capacity, data completeness, and antibody reconstruction accuracy. Furthermore, XA-Novo successfully reconstructs three immunotherapeutic antibodies with unknown sequences, and in vitro assays validating that the generated antibodies exhibit equivalent functionality to their commercial counterparts. Furthermore, XA-Novo excels in discriminating mixtures of COVID-19 neutralizing antibodies, achieving over 99% sequence coverage accuracy, further validating the effectiveness and flexibility of our solution. We anticipate that XA-Novo will be a powerful tool for future antibody research applications.
Introduction
Antibodies are indispensable in modern medicine and research, yet determining their amino acid sequences directly from protein samples—without access to the original DNA—remains a significant challenge. While mass spectrometry (MS) offers a faster alternative to traditional hybridoma technology, it faces three primary hurdles: information gaps due to incomplete digestion, the "isomer trap" where Leucine (Leu) and Isoleucine (Ile) are indistinguishable by mass, and the computational complexity of assembling thousands of short peptide reads into a full-length sequence, especially within complex mixtures.
XA-Novo addresses these challenges by optimizing both the "wet lab" sample preparation and the "dry lab" computational algorithms, providing a high-throughput, accurate, and robust sequencing solution.
Methodology
The XA-Novo pipeline introduces innovations at three critical stages:
SP-MEGD: Single-Pot Multi-Enzymatic Gradient Digestion
Traditional protocols often struggle with sequence gaps. XA-Novo utilizes a novel "gradient digestion" method using five different enzymes, with samples collected at 2, 4, and 6-hour intervals. This approach intentionally generates a gradient of overlapping peptides of varying lengths. These "missed cleavages" provide the essential "head-to-tail" evidence required to assemble the full sequence without gaps.
Deep Learning Peptide Sequencing
The system employs Casanovo, a Transformer-based deep learning model, to translate raw mass spectra into peptide sequences with high precision. Casanovo was trained on the MassIVE KB v1 dataset, containing over 2.1 million precursors, ensuring robust performance across diverse samples.
Fusion Assembler
The Fusion assembler acts as the core computational engine. It utilizes a beam search strategy—a technique borrowed from natural language processing—to explore multiple possible sequence paths simultaneously. By maintaining the top 20 candidates at each step, the assembler avoids errors caused by noisy MS peaks. It also integrates template information from germline databases and utilizes diagnostic "w-ions" to resolve the Leu/Ile ambiguity.
Experimental Results
The authors validated XA-Novo through a series of rigorous benchmarks:
- Superior Accuracy: XA-Novo achieved 100% accuracy in the Complementarity-Determining Regions (CDRs), which are critical for antibody binding. It outperformed industry-standard tools like PEAKS AB and Stitch.
- Blind Testing and Functional Validation: The system successfully reconstructed three commercial therapeutic antibodies with undisclosed sequences (anti-mouse CD4, CD8α, and CSF1R). In vitro assays confirmed that these synthesized antibodies exhibited equivalent functionality to their commercial counterparts in clearing T-cells and macrophages.
- Mixture Deconvolution: In tests involving mixtures of up to three different COVID-19 neutralizing antibodies, XA-Novo maintained over 99% sequence coverage accuracy, demonstrating its ability to distinguish between highly similar proteins.
- Sensitivity: The method proved reliable with as little as 50 µg of purified antibody, making it accessible for low-yield samples.
Discussion and Conclusion
XA-Novo represents a significant advancement in de novo antibody sequencing. By turning "messy" digestion into a feature rather than a bug and applying advanced search strategies to assembly, it overcomes long-standing bottlenecks in the field.
While highly effective, future work will focus on reducing template dependency for non-model organisms and scaling the technology to handle the extreme complexity of polyclonal samples found in raw serum. Currently, the high accuracy of XA-Novo relies on high-resolution MS data, such as that from an Orbitrap Eclipse. To facilitate broader adoption, the team has launched a user-friendly web tool at xa-novo.com.
Summary updated