Transforming Drug Discovery with Quantum Machine Learning

Predicting binding affinity for drug design using classical machine learning faces challenges in capturing the complexity of molecular interactions, representing detailed molecular structures, accounting for quantum effects, handling conformational flexibility, limited availability of labeled training data, and generalizing to novel compounds. These limitations can impact the accuracy and reliability of predictions, highlighting the need for advanced techniques like quantum machine learning to overcome these challenges effectively.

Case Study

Solving for speed and accuracy in drug discovery.

Classical Problem: Drug discovery is a long and expensive process. It takes substantial resources to test the binding affinity between a potential drug and target proteins as well as its effects on patients.

Project Goals: Cut expenses and duration of early drug discovery phases by efficiently predicting the binding affinity between a potential drug and target proteins. As a result, transform drug discovery from slow, high-failure process into rapid, patient-centric model.

Research

Lead Researcher: Laia Domingo, Purdue University Data Mine

Ingenii is developing an end-to-end quantum machine learning pipeline that can better predict the binding affinity between a potential drug and target proteins.

Our pipeline will include:

  • PDB-bind data

  • Hybrid 3DCNN to accelerate the training process

  • Quantum fusion model to improve the accuracy

  • Improve prediction performance and decrease costs

BENEFITS

Enhanced Predictive Accuracy

Quantum neural networks leverage the power of quantum computing and deep learning to achieve superior predictive accuracy in drug binding affinity prediction. By incorporating quantum representations and principles into neural network architectures, these models can capture the intricate quantum interactions and electronic structure details that play a vital role in binding affinity. This leads to more accurate predictions, enabling researchers to prioritize and focus on the most promising drug candidates.

Efficient Exploration of the Chemical Space

Quantum neural networks enable efficient exploration of the vast chemical space involved in drug design. By leveraging quantum computing capabilities, these models can generate and evaluate a wide range of molecular structures and configurations, significantly expanding the search space for potential drug candidates. This ability to efficiently explore chemical space enhances the likelihood of identifying novel compounds with high binding affinity, potentially accelerating the discovery of new therapeutics.

Speed & Efficiency

Quantum neural networks offer the potential for accelerated predictions of binding affinity compared to traditional computational methods. Quantum computing's parallel processing capabilities enable rapid evaluation of complex quantum interactions, reducing the time required for affinity predictions. This efficiency is particularly advantageous in the iterative process of drug design, where timely decisions and optimization are critical.

Reduced Cost & Resources

By accurately predicting binding affinity, quantum neural networks can help streamline the drug design process, reducing the need for extensive experimental testing. This reduction in costly and time-consuming experimental efforts significantly lowers the overall cost of drug development. Additionally, optimizing compound libraries using quantum neural networks can reduce the number of compounds requiring synthesis and testing, further optimizing resource allocation.

Insight into Molecular Interactions

Quantum neural networks provide valuable insights into the underlying molecular interactions that influence binding affinity. By combining quantum representations with neural network architectures, these models can uncover complex relationships between molecular features, atomic interactions, and binding affinities. This understanding of the fundamental mechanisms driving binding affinity aids in rational drug design, allowing researchers to optimize compounds based on specific molecular interactions.

Facilitating Personalized Medicine

Quantum neural networks have the potential to contribute to personalized medicine by predicting binding affinities for specific patient populations. By considering individual genetic variations, quantum neural networks can guide the selection of drug candidates that are likely to exhibit high binding affinity and therapeutic efficacy for specific genetic profiles. This personalized approach increases the likelihood of successful treatment outcomes and reduces the risk of adverse reactions.

PIPELINE

Results

Our research shows the following results:

  • 20% saving in training parameters​

  • 20-40% saving in training times and costs derived from GPU usage

  • ~5% improvement in prediction performance​

  • Better NN convergence than classical fusion models​

Explore our partnership with Purdue

  • Our Chief Science Officer is collaborating with Purdue Data Mine on improving the speed and accuracy of cancer precision therapy. We plan to do this using quantum computing, not only considering molecule and protein interactions on a proteomic level, but also the patient data, resulting in a complex quantum machine learning (QML) model which would not be possible with classical methods and resources.

  • Meet our advisor Dr. Sabre Kais, a lead Director of the Purdue Center for Quantum Technologies.