💊 *Deep Pharma Challenge 2026: Drug-Drug Interaction Prediction*

Can your model save lives by predicting dangerous drug combinations before they are prescribed?

Context
Modern medicine relies on polypharmacy—prescribing multiple drugs to treat complex conditions. However, this comes with a hidden cost: Drug-Drug Interactions (DDIs).

30% of adverse drug events are caused by DDIs.

$30 Billion+ is lost annually in healthcare costs due to preventable interactions.

The Gap: Clinical trials cannot test every possible pair of drugs. We rely on post-market surveillance (Pharmacovigilance) to find these risks, often too late.

The Mission
Your goal is to build a Multi-Task Machine Learning Model that predicts the safety profile of drug pairs. You must predict not just if an interaction occurs, but how severe it is, what specific side effects will happen, and how likely they are compared to the general population.

The Tasks
This is a triple-prediction challenge. For every Drug Pair (A + B), you must output:

Severity Classification: Is the interaction Minor, Moderate, Major, or a Contraindication?

Side Effect Prediction (Binary): Which of the 50 specific adverse events (e.g., Nausea, Arrhythmia) will occur?

Risk Quantification (Regression): What is the Proportional Reporting Ratio (PRR) for each side effect? (A measure of statistical alerting used by the FDA).

The Data
You are provided with a rich, multimodal dataset derived from real FDA Adverse Event Reporting System (FAERS) data:

Chemistry: SMILES strings representing the exact molecular structure of each drug.

Pharmacology: Text descriptions of Mechanisms of Action (MoA), Metabolism, and Transporters.

Scale:

Training Set: ~15,000 confirmed interacting pairs.

Test Set: ~4,000 held-out pairs (Cold Start / New Scaffolds).

Why Join?

Real-World Impact: Your code could be the foundation for next-gen Clinical Decision Support Systems (CDSS).

Scientific Discovery: Help uncover hidden patterns between molecular structure and biological toxicity.