Paper

See https://www.biorxiv.org/content/10.1101/2024.01.25.577287v1.full.pdf

Prefilter

Passing good candidates to the forward/backward algorithm using NNs.

In particular, we aim to emulate HMMER.

Evaluating

To run on custom query and target sequences, edit one of the config/ files with your specifications. Namely, edit the the query_file and target_file parameters to specify your data. The target embedding information will be saved off to the specified target_embeddings, target_names and target_lengths paths to speed up future searches.

To speed up the search, if you have multiple CPUs, you can set num_threads to the number of CPUs, which will parallelize embedding creation and search.

Increasing nprobe will lead to more accurate results, but will be slower.

You can run evaluation with

python3 src/evaluate.py {configfile} where configfile is the name of the config file that you changed.