Reproduce “Diffusion Decoding for Peptide De Novo Sequencing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chi-en Amy Tai, Alexander Wong
This research explores applying diffusion decoding to peptide de novo sequencing, an alternative to traditional autoregressive methods. The study shows that diffusion decoders, particularly when trained with the DINOISER loss function, can achieve a statistically significant improvement in amino acid recall, reaching 0.454 compared to the baseline's 0.081. While individual amino acid prediction improved, peptide-level accuracy was not achieved due to issues with generating overly long, noisy sequences.
Key insights
- Diffusion decoders demonstrate the ability to enhance amino acid recall in peptide sequencing, indicating potential for non-autoregressive sequence generation.
- The DINOISER loss function proved most effective for training discrete diffusion models in this context, leading to higher amino acid recall compared to cross-entropy.
- Integrating knapsack beam search into the diffusion decoding process surprisingly degraded performance and substantially increased computational cost.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.