By integrating 969 ML models across 693 human targets, ProfhEX enables large-scale bioactivity estimation for rapid off-target mapping and safety de-risking.

Modern drug discovery demands a full view of biological interactions. ProfhEX provides an ML-driven compass to navigate this complexity, estimating bioactivity against nearly 700 human targets. It enables researchers to map polypharmacology and identify safety risks or repurposing opportunities before synthesis.

Drug development often fails due to unforeseen "off-target" interactions causing safety risks or poor efficacy. Current computational tools frequently lack the target coverage, updated data, or scalability needed to provide a clear polypharmacological profile during critical early-stage selection.

The Exscalate team expanded ProfhEX (profhex.exscalate.eu), a free web-platform using ML trained on a massive dataset of 5 million curated bioactivity data. Instead of focusing on isolated interactions, ProfhEX characterizes the multiple modes of action of a molecule—including its binding affinity, agonism, and antagonism—across a vast target landscape.

Highlights:

• Unprecedented Library: 969 ML models covering 693 human targets, built on one of the industry's largest curated bioactivity datasets.

• Validated Performance: Within its applicability domain, ProfhEX identifies primary targets with up to 91% accuracy (Top-10), state of the art tools.

• Scientific Reliability: With an average R2 of 0.72 and RMSE of 0.67, the platform’s precision matches experimental lab variability, acting as a reliable "digital twin" for in vitro assays.

ProfhEX turns complex ML inference into a strategic advantage:

• Accelerated De-risking: Early estimation of molecular liabilities allows teams to prioritize compounds with cleaner safety profiles, reducing late-stage attrition.

• Evidence-Based Repurposing: Vast target coverage uncovers unexpected bioactivities, identifying new therapeutic indications for existing scaffolds.

• Holistic Lead Optimization: Beyond single-protein focus, researchers can visualize a molecule’s full polypharmacological fingerprint, ensuring superior selectivity and fewer side effects.

Accuracy is highest within the platform's "applicability domain." Future updates will expand coverage to underrepresented target families and integrate ADMET properties with metabolic pathways. The team is also implementing "uncertainty quantification" to provide confidence scores for every assessment.

ProfhEX provides a robust, evidence-based foundation for mapping polypharmacology in early discovery. For a deep dive into the ML architecture and performance benchmarks, read the full Application Note in the Journal of Chemical Information and Modeling.