Quantum Machine Learning for Drug Discovery
Improving molecule generation using classical and quantum techniques
Jose-Ignazio Latorre, National Quantum Computing Hub (NQCH)
Alessandro Luongo, Centre for Quantum Technologies (CQT) and National University of Singapore (NUS);
Enrico Petretto, Centre for Computational Biology (CCB)
Jacques Behmoaras, Centre for Computational Biology (CCB)
The goal of this research proposal is to leverage classical and quantum computation to enhance the process of generating molecular descriptions for potential drug candidates. Existing methodologies, such as MolGAN and QMolGAN leverage on the QM9 database, which consists in the ~134k smallest organic molecules containing up to 9 heavy atoms (C, O, N, or F; excluding H) along with their quantum properties. These methods exhibit limitations in the quality and the size of the molecules generated, and size of the molecular databases. This research aims to overcome these challenges with the first objective of implementing quantum neural networks (QNNs) and, inspired from QMolGAN, to improve quantum generative adversarial networks (qGAN) for better molecule generation. In a second phase, these methods will be tailored towards generation of small molecules that can inhibit WWP2, an antifibrotic drug-target[1,2]; the best molecules will be synthesized and tested experimentally for physical chemistry and antifibrotic properties.
References
[1] H Chen, A Moreno-Moral, F Pesce, N Devapragash, M Mancini, E Ling Heng, M Rotival, P K Srivastava, N Harmston, K Shkura, OJL Rackham, W-P Yu, Xi-Ming Sun, N Gui Zhen Tee, E Tan, PJR Barton, LE Felkin, E Lara-Pezzi, G Angelini, C Beltrami, M Pravenec, S Schafer, L Bottolo, N Hubner, Costanza Emanueli, SA Cook & E Petretto. WWP2 regulates pathological cardiac fibrosis by modulating SMAD2 signaling. Nature Communications 2019; 10(1):3616, PMID: 31399586 7
[2] H Chen, G Chew, N Devapragash, JZ Loh, KY Huang, J Guo, S Liu, ELS Tan, NGZ Tee, MM Mia, MK Singh, S Chen, A Zhang, J Behmoaras & E Petretto. The E3 ubiquitin ligase WWP2 regulates pro-fibrogenic monocyte infiltration and activity in heart fibrosis. Nature Communications 2022; 13(1):7375, PMID: 36450710
Quantum Inspired Algorithms for Gene Regulatory Network Studies
Apply methods developed for the study of quantum systems to gather insights on the gene regulatory network at single cell level
Jose-Ignazio Latorre, National Quantum Computing Hub (NQCH)
Dario Poletti, Singapore University of Technology and Design (SUTD) and Centre for Quantum Technologies (CQT)
Enrico Petretto, Centre for Computational Biology (CCB)
Jacques Behmoaras, Centre for Computational Biology (CCB)
The power and the course of quantum mechanics is that to describe classically a quantum system one needs a memory that scales exponentially with the system size. This is extremely powerful because one can capture significant more information in a quantum system. The problem is that without quantum computers and simulators, one would need a classical computer with an exponentially large memory to describe exactly any possible quantum state. However, one may not need an exact representation of quantum states as sometimes approximate solutions can be extremely insightful. Following this logic, a number of approximate methods has been developed to tackle complex quantum systems. Extracting useful information from complex scenarios is not exclusive to the study of quantum systems. For example, biological systems are complex: multiple agents can produce different outcomes, however the agents tend to work cooperatively, and depending on the group of agents they work with, sometimes overlapping, different dynamics can emerge. It is an extremely demanding task to figure out what each combination can produce and how the overall system would react to other different stimuli. One of these scenarios is the of the gene regulatory network. With this project we intend to apply methods developed for the study of quantum systems (tensor networks), which run on classical computers, to gather insights on the gene regulatory network using real, yet anonymized, data.