Quantum Generative Models for Small Molecule Drug Discovery (2101.03438v1)

Abstract: Existing drug discovery pipelines take 5-10 years and cost billions of dollars. Computational approaches aim to sample from regions of the whole molecular and solid-state compounds called chemical space which could be on the order of 1060 . Deep generative models can model the underlying probability distribution of both the physical structures and property of drugs and relate them nonlinearly. By exploiting patterns in massive datasets, these models can distill salient features that characterize the molecules. Generative Adversarial Networks (GANs) discover drug candidates by generating molecular structures that obey chemical and physical properties and show affinity towards binding with the receptor for a target disease. However, classical GANs cannot explore certain regions of the chemical space and suffer from curse-of-dimensionality. A full quantum GAN may require more than 90 qubits even to generate QM9-like small molecules. We propose a qubit-efficient quantum GAN with a hybrid generator (QGAN-HG) to learn richer representation of molecules via searching exponentially large chemical space with few qubits more efficiently than classical GAN. The QGANHG model is composed of a hybrid quantum generator that supports various number of qubits and quantum circuit layers, and, a classical discriminator. QGAN-HG with only 14.93% retained parameters can learn molecular distribution as efficiently as classical counterpart. The QGAN-HG variation with patched circuits considerably accelerates our standard QGANHG training process and avoids potential gradient vanishing issue of deep neural networks. Code is available on GitHub https://github.com/jundeli/quantum-gan.

• The paper presents QGAN-HG, a hybrid quantum-classical GAN that reduces parameter overhead while efficiently exploring expansive chemical space.
• It employs a variational quantum circuit for feature extraction, achieving a 14.93% parameter reduction compared to classical methods.
• The use of patched circuits improves training stability, enhancing molecule generation quality and speeding up drug discovery pipelines.

Quantum Generative Models for Small Molecule Drug Discovery

The paper under review focuses on advancing drug discovery methodologies through the employment of quantum generative models, specifically targeting small molecule compounds. Traditional drug discovery pipelines are protracted and financially intensive, necessitating innovative computational strategies to explore the vast chemical space, which is often estimated to contain as many as $10^{60}$ potential compounds. In this context, the authors propose a quantum generative adversarial network (QGAN) framework, designed to outperform classical approaches under constraints of qubit resource availability.

The core proposition within this paper is the introduction of a hybrid quantum GAN architecture, named QGAN-HG. This model capitalizes on the expressiveness of quantum computing to navigate the exponentially large chemical space using fewer computational resources than would typically be necessary. The QGAN-HG is structured with a quantum component acting as a generator, which is paired with a classical discriminator. This configuration facilitates enriched molecule representation learning, thus implicating quantum computing's aptitude for complex data distributions in the drug discovery process.

Specifically, the quantum generator includes a variational quantum circuit, which is capable of extracting meaningful feature vectors from reduced quantum states. This is crucial, as the authors highlight that classical GAN frameworks are limited in their capacity to cover the expanse of chemical space due to parameter overhead and sampling inefficiencies. Notably, the QGAN-HG with a 14.93% reduced parameter set demonstrates the capacity to learn molecular distributions with efficiency rivalling its classical counterparts, an impressive accomplishment underscored by the minimal resources utilized.

The paper extends the framework through the implementation of patched circuits, a methodological variant designed to mitigate performance degradation associated with deep network training, such as gradient vanishing. The QGAN-HG and its patched variant significantly optimize training time while preserving, and at times, enhancing drug molecule generation quality, as evaluated through Fréchet distance metrics and RDKit drug property scores.

It is essential to address the implications of these findings. The proposed QGAN-HG model not only emboldens the use of quantum machine learning in chemistry but potentially accelerates the iterative processes involved in drug synthesis and validation. The application can streamline transitions from molecular design to preclinical evaluations by generating viable candidate libraries more efficiently.

Prospectively, further developments in this area could center on integrating these advancements within actual quantum computing environments. As hardware constraints relax with technological advancement, executing inferences and training on real quantum devices could yield even richer molecular data sets and conformational properties. The accuracy and efficacy of drug discovery processes could profoundly benefit from such developments, thereby shortening timelines and reducing costs in bringing new compounds to market.

In conclusion, the engagement of quantum generative models in drug discovery embodies a promising intersection of computational chemistry and quantum computing. The approaches detailed in this paper highlight potential transformative gains in efficiency and capability, setting a compelling precedent for future exploration and application of quantum-enhanced computational models in the biopharmaceutical domain.