Bedhost-ui: Implement new text search interface

[donaldcampbelljr]

Implement new text search interface on the bed base landing page. Right now the current one is Proof of Concept and needs to be functional and updated to use dedicated endpoints for searching.

UI needs to be updated to support text and bedfiles. One search endpoint for both is suggested.

Some Steps:

Implement /search endpoint :

1. [x] transform user input into a vector with a sentence transformer
2. [x] convert the vector into a region set vector
3. [x] use a K-nn search to find nearby region set vectors
4. [x] Return those as a list to the user, with links to the BED files.

Implementation details:

• [x] The k-nn part should use the EmSearchInterface and be backed by a qdrant instance
• [x] The sentence transformer and region set part should be handled by a Text2BedNN Python class, with capability of converting convert from sentence embedding to a BED file embedding using trained models.

Implement a user interface to /search:

• [x] it should just be a simple web form that POSTs the search text to /search, and then renders the results.

[nsheff]

To capture the BED-to-BED search, I've created a new issue: https://github.com/databio/bedhost/issues/94

This can focus on TEXT-to-BED

[nsheff]

Related to: https://github.com/databio/bedbase/issues/37

[nsheff]

Duplicate of databio/bedbase#31 and databio/bedbase#8

[nsheff]

TextToBEDNN class:

• sentence transformer (SentenceTransformer)
• NN model (tf.keras.models.Sequential())
• Region2vec ExModel (Exmodel, contains: 1) tokenizer; 2) gensim model))
• search interface (EmSearchInterface object)

[ClaudeHu]

The newest vec2vec model is on huggingface, can be installed by:

from geniml.text2bednn import Vec2VecFNN

model = Vec2VecFNN("databio/v2v-geo-hg38")

[khoroshevskyi]

@ClaudeHu is this issue solved?

[khoroshevskyi]

text to search interface was added in v0.4.0