🌱 V Hybrid Predictor

Inspiration

There wasn’t really any project that inspired this idea — it came from realizing how little attention is given to the challenges agricultural scientists, biologists, and genetic researchers face while developing new hybrid plant species. I felt a strong need for a tool that could help them predict hybrid viability, improve efficiency, and save years of trial and error. V Hybrid Predictor was born from that gap — a practical AI solution to assist real research.

What it does

V Hybrid Predictor is an AI-powered system that predicts whether two plant varieties can successfully hybridize or not. It analyzes detailed features such as fruit traits, genetic similarity, environmental tolerance, yield, and disease resistance to estimate hybrid viability percentage, expected yield, market value, and even flavor potential. This helps scientists and breeders make smarter decisions before performing real-world hybridization.

How we built it

We started by creating a rich dataset of tomato varieties containing over 40+ biological and environmental features. Then, we generated pairwise combinations of potential parent varieties and applied genetic logic rules to calculate compatibility. The model was trained using Artificial Neural Networks (ANN) optimized with Gradient Descent Optimization to capture hidden relationships between parent traits and hybrid outcomes. We also included data preprocessing, feature encoding, and normalization to ensure the model learns efficiently.

Challenges we ran into

• Finding real and consistent plant trait data was tough.
• Translating biological compatibility into machine-readable numerical form was tricky.
• Ensuring the model didn’t overfit while maintaining accuracy.
• Designing features like genetic distance and hybrid viability score to reflect realistic biological conditions.

Accomplishments that we're proud of

• Created a complete AI-based hybrid prediction system from scratch.
• Simulated real-world genetics and environmental traits without external datasets.
• Built a working dataset of 20,000+ hybrid combinations.
• Designed a simple yet effective deep learning model that can evolve for other crops.

What we learned

• How AI can truly assist biological research and plant breeding.
• The importance of feature engineering and realistic domain knowledge in model performance.
• That blending biology and machine learning opens a new direction for sustainable agriculture research.

What's next for V Hybrid Predictor

We plan to:

• Integrate real genotyping and SNP data for higher accuracy.
• Expand the model to other crop species beyond tomatoes.
• Create a Streamlit-based research tool for scientists to test hybrid predictions interactively.
• Partner with agricultural institutions to collect real hybridization data and refine predictions.

Built With

• ann
• matplotlib
• numpy
• python
• pytorch
• scikit-learn
• seaborn
• streamlit