Reproduce “Blind Biological Sequence Denoising with Self-Supervised Set Learning” 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
Nathan Ng, Ji Won Park, Jae Hyeon Lee, Ryan Lewis Kelly, Stephen Ra, Kyunghyun Cho
This research introduces Self-Supervised Set Learning (SSSL), a neural network framework designed for blindly denoising biological sequences from noisy long-read data without requiring ground-truth sequences. SSSL effectively reconstructs accurate sequences, particularly from small sets of subreads, by learning an implicit consensus representation and demonstrating improved error reduction on both simulated and real antibody sequencing data.
Key insights
- Learning an implicit, generalized representation of a true source sequence from a set of noisy observations, without explicit ground-truth labels, is an effective strategy for biological sequence denoising.
- The proposed "midpoint loss" and Kernelized Maximum Mean Discrepancy (MMD) in the latent space are critical for guiding the set aggregator to produce a robust consensus embedding from heterogeneous noisy subreads.
- Novel self-supervised metrics, namely Leave-One-Out (LOO) Edit Distance and Fractal Entropy, provide reliable and complementary methods for evaluating blind denoising performance in the absence of ground-truth sequences.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.