UbiPredict: AI-based Ubiquitination Site Predictor

• 

  Two-stage deep learning pipeline for ubiquitination and localization prediction.

Inspiration

Protein ubiquitination plays a vital role in cell regulation and disease progression, yet experimental detection is costly and time-consuming. We were inspired to create an AI-driven solution that accelerates protein analysis for researchers.

What It Does

UbiPredict is a two-stage deep learning framework that predicts protein ubiquitination sites and subcellular localization using ESM-based embeddings and CNN architectures. It enables accurate, automated protein annotation and supports modern bioinformatics workflows.

How We Built It

We integrated Python, PyTorch, and ESM protein language models for feature extraction and prediction. The system was trained on curated datasets and validated using performance metrics like precision, recall, and F1-score. A lightweight web interface allows users to upload sequences and visualize predictions.

Challenges

We faced challenges with large dataset handling, GPU optimization, and model integration between stages. Managing data storage and version control for deep learning models was also a significant hurdle.

Accomplishments & Learning

We successfully built an end-to-end predictor achieving high accuracy, learning advanced model optimization, and mastering bioinformatics–AI integration.

Built With

• esm
• flask
• matplotlib
• numpy
• pandas
• python
• pytorch
• scikit-learn