yinboliu-git

yinboliu-git / MoEST

Public

A deep learning framework for predicting gene expression from 3D spatial transcriptomics data using morphology-guided mixture of experts.

33
3
100% credibility
Found Feb 05, 2026 at 20 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A research tool for predicting gene expression levels in 3D spatial transcriptomics data from histopathology images and coordinates using a morphology-guided approach.

How It Works

1
🔍 Discover 3DMoEST

You stumble upon this clever tool that helps predict gene activity in 3D maps of body tissues using everyday microscope pictures.

2
📦 Gather your tissue info

Collect your pictures of tissue slices, measurements of active genes, and notes on where each spot is located in 3D space.

3
⬇️ Pick up the image expert

Grab the ready-made helper that understands tissue pictures deeply from years of looking at millions of them.

4
🔬 Reveal tissue secrets

Let the helper scan your pictures to pull out smart clues about shapes and patterns in the tissues.

5
🏗️ Build your gene predictor

Feed in your clues, locations, and real gene data to train a personal guesser that learns the patterns.

6
🔮 Forecast gene activity

Point it at new spots in your 3D tissue and watch it predict which genes are busy there.

🎉 See biology come alive

Marvel at heatmaps and overlays where predicted gene levels match real measurements, unlocking new insights into your tissues.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 20 to 33 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is MoEST?

MoEST is a Python deep learning framework that predicts gene expression from 3D spatial transcriptomics data—histopathology images, 3D coordinates, and Sobel-derived morphology gradients. It tackles the high cost of sequencing by imputing full expression profiles using a morphology-guided mixture of experts, trained on datasets like HER2-ST, MISAR, and openST. Users run quickstart scripts to extract UNI features, train via k-fold or single splits, and generate inference heatmaps, expert routing maps, and gradient visuals from H5 inputs.

Why is it gaining traction?

Unlike vanilla deep learning vs machine learning baselines, MoEST's sparse experts auto-route based on tissue edges, enforcing spatial smoothness with coupled losses and Fourier 3D encoding—yielding robust predictions on count data via negative binomial modeling. Deep learning with python users love the seamless UNI integration (ViT-L/16 pathology foundation model) and masked training for noisy visuals. It's a github deep learning arcgis-style pipeline: prep data, train on A100s, visualize results fast.

Who should use this?

Computational pathologists modeling 3D tissue like mouse embryos (MISAR) or breast cancer (HER2-ST), needing expression imputation without extra sequencing. Deep learning deutsch/ki researchers in spatial omics prototyping morphology-aware nets. Bioinformaticians extending scvi/scanpy workflows for generative ki predictions from images alone.

Verdict

Solid prototype for deep learning goodfellow-inspired experts in pathology—installs cleanly, runs end-to-end—but 33 stars and 1.0% credibility signal early maturity with thin docs and no tests. Fork if you're in deep learning ai for ST; otherwise, monitor for polish.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.