🧬 GeNLP

GeNLP: An interactive web application for microbial gene exploration and prediction

🌐 Visit the GeNLP website

The quick user guide is available at our GitHub wiki

This repository contains the implementation of GeNLP, a user-friendly web server to explore gene relationships!
The server is based on a pre-trained published language model:
"Deciphering microbial gene function using natural language processing"

Weights and trained model are available on the paper's GitHub repository.

Getting Started

This web service does not require any previous installation.
In the following use case, we demonstrate the core features of the web server together with potential biological insights.
Upon entering the web service, a map of all genes supported by our model is presented, color-coded by their functional group, unknown proteins are colored in light grey.

Search for sequence

We start by using the server on an unknown protein sequence, with no mapping to existing databases.

>Protein1
MTKLELLVSVMVDGKWYSTDDLVSRVGHRFSATKHVAEKQGYQFEKRREGMRFEYRMVST
TIELAR

To obtain a prediction, go to the search panel and choose Sequence. The app will run in the background and find the matching gene identifier in our model for Protein1.
NOTICE: This process might take a few minutes 🤗

After the search is completed, the predicted gene identifier and its corresponding prediction scores will be available for download. We will use the predicted gene identifier: hypo.clst.15503442 to explore Protein1.

Explore prediction

note: if you ran the search by sequence the predicted gene hypo.clst.15503442 will be highlighted in the genomic map.
Go to the search bar and select model word mode. Type the identifier name in the search bar below (it should auto-complete).
The highlighted dot on the space marked hypo.clst.15503442. By zooming in to the cluster in which this dot resides we can interactively inspect the neighboring genes, most of which are related to the CRISPR-Cas system. For in-depth analysis, click on the dot to open a panel with the following information:

• Predicted class
• Prediction confidence
• Neighbors
• Gene prediction
• NCBI NR description
• Gene count in family

Gene function prediction

The predicted class is Prokaryotic Defense System and the prediction is reliable (trusted prediction is True).
By clicking on FUNC PRED a panel will open showing all scores received by the model for the different classes. Here, the Prokaryotic Defense System prediction was unequivocal, with a score close to 1 (possible scores are between 0 to 1).

Analyzing neighbors

By clicking on the NEIGHBORS tab a bar plot will be opened with the closest 10 neighbors in the space. The neighbors are:

1. Five Cas proteins, including Cas3, Csc1, Csc2 and two variants on Csc3.
2. Five Hypothetical proteins, all with a trusted prediction of a "Prokaryotic Defense System". Hovering on a specific bar will show extended information on each gene.
   

This strengthens the confidence that this protein is related to Defense mechanisms, and more specifically to Subtype I in which Csc proteins are apparent.

Note: distances were calculated on a high dimensional space, thus elements might not seem to be close in the 2D space map.

Taxonomic mapping of genes in a family

Selecting the TAX MAP tab will display a bar plot showcasing the top 10 taxonomic orders associated with the gene family hypo.clst.15503442.
The plot also indicates the percentage of genes from this family that have a known taxonomy in the database—specifically, in this case, 79.17% of genes were mapped.