LGTM#

Code for LGTM: Gaussian Process Modulated Neural Topic Modeling for Longitudinal Microbiome.

Live demo: https://lgtm-web.streamlit.app/

Contact: Xiao Yuan (yuanx749@gmail.com).

Installation#

Linux is recommended.

Clone this repository and enter the project folder.

Install the package with uv:

uv venv .venv
source .venv/bin/activate
uv pip install -e .

For development, create and activate the environment with Mamba:

mamba env create -p ./env -f env-dev.yml
mamba activate ./env

For experiments with other methods, use env-exp.yml.

Usage#

As a demo, first download the public HMP2 dataset, then open examples/hmp_gp.ipynb for training and visualization.

chmod +x scripts/hmp_download.sh
./scripts/hmp_download.sh

Python API:

import pandas as pd
from lgtm import LGTM, LGTMConfig

metadata = pd.read_csv("metadata.csv")
microbiome = pd.read_csv("microbiome.csv")

config = LGTMConfig(
    latent_dim=6,
    n_epoch=100,
    batch_size=64,
    learning_rate=0.05,
    seed=42,
)
model = LGTM(config).fit(metadata, microbiome)

sample_topic = model.sample_topic_
topic_taxon = model.topic_taxon_

fig, axes = model.plot_si()
fig, axes = model.plot_topics()
fig, axes = model.plot_gp(topic=1)