Transformer-based Objective-reinforced Generative Adversarial Network to Generate Desired Molecules

Abstract

Monolithic software encapsulates all functional capabilities into a single deployable unit. But managing it becomes harder as the demand for new functionalities grow. Microservice architecture is seen as an alternative as it advocates building an application through a set of loosely coupled small services wherein each service owns a single functional responsibility. But the challenges associated with the separation of functional modules, slows down the migration of a monolithic code into microservices. In this work, we propose a representation learning based solution to tackle this problem. We use a heterogeneous graph to jointly represent software artifacts (like programs and resources) and the different relationships they share (function calls, inheritance, etc.), and perform a constraint-based clustering through a novel heterogeneous graph neural network. Experimental studies show that our approach is effective on monoliths of different types.

Artificial Intelligence for Surface‐Enhanced Raman Spectroscopy

“…Predict optical properties GCN [88] SMILES Design new valid molecules VAE, [ 65,[89][90][91] GAN, [65,[92][93][94][95][96] RL [65] Selected molecular descriptors Predict the maximum absorption wavelength…”

“…TransORGAN [94] A model consisting of two GANs to translate transcriptomic profiles from different organs, sexes, and age groups in rodents.…”

“…GAN, as mentioned in the above section, can also control the design of new molecules by using the generator network. [ 65 ] Some variants of GAN in Table 3 including ORGAN, [ 92 ] ORGANIC, [ 93 ] TransORGAN, [ 94 ] LatentGAN, [ 95 ] and MolFilterGAN, [ 96 ] have been proposed to accelerate the target molecule discovery integrated with the property evaluation (Figure 3d).…”

GxVAEs: Two Joint VAEs Generate Hit Molecules from Gene Expression Profiles

“…The power of deep learning and computational chemistry enables the generation of new molecules with desired bioactivity for specific therapeutic targets. Deep generative models, such as generative adversarial networks (GANs) (Guimaraes et al 2017;De Cao and Kipf 2018;Li et al 2022) and variational autoencoders (VAEs) (Oliveira, Da Silva, and Quiles 2022;Dollar et al 2021;Kusner, Paige, and Hernández-Lobato 2017;Jin, Barzilay, and Jaakkola 2018) accelerate the drug discovery process by employing computational models to generate molecules having a specific bioactivity. Such models learn the underlying patterns and relationships within molecular structures by training on datasets of known molecules with the desired properties.…”

De Novo Generation of Chemical Structures of Inhibitor and Activator Candidates for Therapeutic Target Proteins by a Transformer-Based Variational Autoencoder and Bayesian Optimization

“…In addressing this issue, deep generative models (i.e., structure generators) for de novo drug design strategies have been actively developed. There have been many previous studies that use the variational autoencoder (VAE) − and generative adversarial network (GAN). − However, most of the previously developed structure generators are chemistry-centric approaches; they focused on generating chemically valid structures, and few considered the comprehensive biological responses caused by interactions between the drug candidate molecules and target proteins. The goal of most previous methods was to optimize properties that could be calculated directly from chemical structures (e.g., the quantitative estimate of drug-likeness and the synthetic accessibility score) or to improve bioactivity by using quantitative structure–activity relationship (QSAR) models in the molecular generation process.…”