Hybrid Quantum-Classical Model for Molecular Generation: Integration of a QCBM and LSTM to Identify Novel Ligands for A2a Receptor

Overview

Drug discovery is a challenging, expensive, and time -consuming process often spanning a decade and requiring an investment of at least $2.5 billion. Further, Drug discovery targeting complex proteins like the A2a adenosine receptor which is related to neurodegenerative diseases such as Parkinson’s is challenging because of the vast chemical space of drug -like molecules, often theorized as 10 60 molecules. This study hypothesized that a hybrid quantum -classical approach, integrating a Quantum Circuit-Born Machine (QCBM) and a Long Short-Term Memory (LSTM) network, could effectively model molecular distributions and generate novel molecules with high binding affinity for the A2a receptor. This model first expanded a dataset of 10 known inhibitors with the STONED SELFIES algorithm, randomly mutating a training set of molecules while preserving their chemical validity. These molecules were converted into their binary Morgan Fingerprint, representing a high dimensionality probability distribution. The QCBM then learned from this data and represented this distribution as pure quantum states, which was the input to an iterative LSTM that generated novel compounds based on the features extracted f rom the QCBM. Finally, a docking simulation was used to produce the binding affinity of the generated molecules, and the three best performers were presented as the results. In conclusion, this work establishes a novel methodology that combines the advantages of classical and quantum computing, successfully generating a number of molecules with high binding affinity to the A2a receptor. By accelerating drug discovery pipelines , such approaches could address unmet medical needs, improving outcomes for millions lacking effective treatments.