A hybrid C++/Python pipeline for remote protein homology detection, coupling ESM-2 language model embeddings with custom Neural-LSH for scalable approximate nearest neighbor search.
-
Updated
Jan 30, 2026 - Python
A hybrid C++/Python pipeline for remote protein homology detection, coupling ESM-2 language model embeddings with custom Neural-LSH for scalable approximate nearest neighbor search.
Bioinformatics capstone of the UoA DI 'Software Development for Algorithmic Problems' trilogy. Protein remote-homolog search on SwissProt-50k via ESM-2 embeddings + 6 ANN methods (brute, LSH, Hypercube, IVF-Flat, IVF-PQ, Neural LSH), benchmarked against BLAST.
Implementation of Neural LSH (Dong et al., ICLR 2020) for ANN search: C++ k-NN graph + KaHIP graph partitioning + PyTorch MLP for learned bucket assignment, with Bayesian HPO via Optuna. Beats HW1's IVF-Flat ~4× at high recall on MNIST.
Add a description, image, and links to the neural-lsh topic page so that developers can more easily learn about it.
To associate your repository with the neural-lsh topic, visit your repo's landing page and select "manage topics."