pnpnpn/dna2vec

Employs a multi-k approach to train embeddings across variable k-mer lengths (k=3 to k=8) simultaneously, capturing sequence patterns at multiple scales. Implements Word2Vec-style distributed representations accessible via a `MultiKModel` interface that supports vector lookup and cosine similarity queries. Includes pretrained models on human genome (hg38) and provides YAML-based configuration for training on custom genomic FASTA datasets.

192 stars. No commits in the last 6 months.