PestiSynth

dashboard
UI
report
UI
architecture workflow

Inspiration

Modern drug discovery has been transformed by machine learning , yet pesticide discovery, which is essentially drug discovery for pests, still relies heavily on slow, expensive wet-lab trials. We saw a gap: researchers, agritech startups, and even students lack accessible computational tools to rapidly test protein–ligand affinity without needing deep biotech expertise.

So we asked: “What if a small student team with no biology background could build a mini computational wet-lab powered entirely by AI?”

This was our “aha” moment , and PestiSynth was born.

What it does

PestiSynth is an AI-driven protein–ligand affinity predictor that helps users:

Select a target pest
Evaluate up to 5 pesticide molecules
Get predicted binding affinity using the PLAPT transformer model
Run a toxicity safety scan using chemical similarity
View a complete Affinity + Safety report with charts & interpretation

simply - It predicts how well a pesticide binds to a pest’s protein and whether it's similar to banned or harmful chemicals. What normally takes wet-labs days or weeks is now accessible in seconds.

How we built it

We combined multiple domains : biology, chemistry, ML, and UI/UX - into one coherent system:

Backend

FastAPI server in Python
Integrated PLAPT (Protein-Ligand Affinity Prediction Transformer)
Custom pipeline for canonical SMILES handling and validation
Safety module using RDKit molecular fingerprints
Local database of banned/hazardous pesticides for similarity matching
- Batched scoring endpoint for multi-molecule evaluation

Frontend

React + modern UI styling
Dynamic pest & pesticide selection grids
Beautiful report generation
Bar charts + scatter plots for affinity visualization
Image gallery for molecular and pest icons
Smooth state management and fast interactions

Data Engineering

Compiled FASTA sequences for 8+ pests from NCBI Canonicalized 20+ pesticide SMILES from PubChem ( supports all chemical compunds input) Created a structured JSON library for quick API access

Challenges we ran into

Zero biology background

We had to learn:

FASTA sequences
Protein structure relevance
Ligand SMILES representations
Pesticide chemical classes
How affinity (µM) relates to binding strength

This was both the hardest and most rewarding part.

Working with a research-grade ML model

PLAPT has specific expectations for:

Input sequence formatting
Canonical SMILES
File paths and ONNX inference

Interfacing everything smoothly took careful debugging.

Making science visually intuitive

We wanted PestiSynth to feel like a real mini drug-discovery app:

clean affinity charts
pest hero section
molecular cards with SMILES tags
accessible explanations for non-experts

Achieving this polish in limited time was a major UI/UX challenge.

Accomplishments that we're proud of

In just one weekend, we integrated real protein FASTA data, SMILES chemistry, and a transformer model to simulate a virtual wet lab. We’re proud that we turned unfamiliar biotech concepts into a functional, end-to-end system that actually works.

We designed a clean UI with molecule cards, rankings, charts, and safety flags so even non-experts can run affinity predictions. Making advanced computational chemistry accessible and intuitive is our biggest accomplishment.

What we learned

Integrating ML models with real biological data
Parsing FASTA protein sequences
Working with SMILES and cheminformatics libraries
Full-stack engineering under time pressure
Designing for clarity in a domain most people find intimidating
How wet-lab and web-lab workflows connect in real research
Communicating complex science in simple visuals

Most importantly: you don’t need a biology degree to build biotech tools , just curiosity and persistence.

What's next for PestiSynth

We plan to extend PestiSynth into a fully QSPR-enabled molecular design platform:

De novo molecule generation (design optimized pesticides)
ADMET prediction pipeline (toxicity, stability, solubility)
3D docking visualization
Custom protein uploads for any organism
Larger safety database (EPA, REACH, FAO datasets)
Enterprise dashboard for agritech companies
Integration with wet-lab robotics for automated testing

Our vision is to help bridge the gap between computational prediction and experimental validation — empowering researchers, students, and agritech innovators.