[Paper Review] CardioGenAI: A Machine Learning-Based Framework for Re-Engineering Drugs for Reduced hERG Liability

Gregory W. Kyro, Matthew T. Martin+2 more|arXiv (Cornell University)|Mar 12, 2024

Cardiac pacing and defibrillation studiesMedicine3 citations

TL;DR

CardioGenAI is a machine learning framework that re-engineers drugs to reduce hERG liability while preserving on-target potency by predicting hERG, NaV1.5, and CaV1.2 channel activities. It successfully redesigned pimozide into fluspirilene, which shows over 700-fold weaker hERG binding, demonstrating its potential to rescue drug candidates stalled due to cardiac safety concerns.

ABSTRACT

The link between in vitro hERG ion channel inhibition and subsequent in vivo QT interval prolongation, a critical risk factor for the development of arrythmias such as Torsade de Pointes, is so well established that in vitro hERG activity alone is often sufficient to end the development of an otherwise promising drug candidate. It is therefore of tremendous interest to develop advanced methods for identifying hERG-active compounds in the early stages of drug development, as well as for proposing redesigned compounds with reduced hERG liability and preserved on-target potency. In this work, we present CardioGenAI, a machine learning-based framework for re-engineering both developmental and commercially available drugs for reduced hERG activity while preserving their pharmacological activity. The framework incorporates novel state-of-the-art discriminative models for predicting hERG channel activity, as well as activity against the voltage-gated NaV1.5 and CaV1.2 channels due to their potential implications in modulating the arrhythmogenic potential induced by hERG channel blockade. We applied the complete framework to pimozide, an FDA-approved antipsychotic agent that demonstrates high affinity to the hERG channel, and generated 100 refined candidates. Remarkably, among the candidates is fluspirilene, a compound which is of the same class of drugs (diphenylmethanes) as pimozide and therefore has similar pharmacological activity, yet exhibits over 700-fold weaker binding to hERG. We envision that this method can effectively be applied to developmental compounds exhibiting hERG liabilities to provide a means of rescuing drug development programs that have stalled due to hERG-related safety concerns. We have made all of our software open-source to facilitate integration of the CardioGenAI framework for molecular hypothesis generation into drug discovery workflows.

Motivation & Objective

• To address the high attrition rate in drug development caused by hERG-related cardiotoxicity.
• To develop a machine learning framework capable of re-engineering existing drugs to reduce hERG channel inhibition.
• To maintain on-target pharmacological activity while minimizing off-target ion channel effects.
• To provide a practical, open-source tool for integrating molecular hypothesis generation into drug discovery workflows.

Proposed method

• The framework employs state-of-the-art discriminative models to predict hERG, NaV1.5, and CaV1.2 channel inhibition.
• It uses generative machine learning to propose structural modifications to existing drug molecules.
• The method incorporates multi-channel activity prediction to assess arrhythmogenic risk beyond hERG alone.
• It applies molecular re-engineering to FDA-approved and developmental compounds, focusing on structural analogs.
• The framework is trained on curated datasets linking chemical structure to ion channel activity.
• All software is open-source to enable integration into industrial and academic drug discovery pipelines.

Experimental results

Research questions

• RQ1 Can machine learning accurately predict hERG inhibition and related off-target ion channel effects for drug candidates?
• RQ2 Can generative models produce structurally modified compounds with reduced hERG affinity while preserving pharmacological activity?
• RQ3 Can the framework rescue drug development programs stalled due to hERG liability?
• RQ4 To what extent does the redesigned compound fluspirilene retain on-target activity while exhibiting drastically reduced hERG binding?
• RQ5 How scalable and generalizable is the framework across different drug classes?

Key findings

• The framework successfully re-engineered pimozide into fluspirilene, a known antipsychotic with over 700-fold lower hERG binding affinity.
• Fluspirilene, a structural analog of pimozide, maintains similar pharmacological activity despite dramatically reduced hERG inhibition.
• The framework's predictive models for hERG, NaV1.5, and CaV1.2 channels demonstrated high accuracy in classifying compound activity.
• Among 100 generated candidates, fluspirilene emerged as a top-ranked compound with optimal safety and activity profile.
• The open-source release of CardioGenAI enables broad adoption and integration into drug discovery workflows for early safety optimization.