Please use this identifier to cite or link to this item:
Title: De novo design of small molecules using variational and conditional variational autoencoders
Author: Zavodnik, Ŝpela
Director/Tutor: Vitrià i Marca, Jordi
Keywords: Aprenentatge automàtic
Xarxes neuronals (Informàtica)
Tesis de màster
Machine learning
Neural networks (Computer science)
Masters theses
Issue Date: 2-Sep-2019
Abstract: [en] Chemical space is estimated to contain over 10 60 small synthetically feasible molecules and so far only a fraction of the space has been explored. Experimental techniques are time-consuming and expensive so computational methods, such as machine learning, are needed for efficient exploration. Here we looked at generative models, more specifically variational autoencoder (VAE) and conditional variational autoencoder (CVAE), used for designing new molecules. In the first part, we evaluated already written VAE and in the second part, we upgraded it to the CVAE. For the conditional vectors in CVAE we used B4 Signatures generated from Chemical Checker describing molecular properties. Both models performed well, however, CVAE showed many advantages.
Note: Treballs finals del Màster de Fonaments de Ciència de Dades, Facultat de matemàtiques, Universitat de Barcelona, Any: 2019, Tutor: Jordi Vitrià i Marca
Appears in Collections:Màster Oficial - Fonaments de la Ciència de Dades

Files in This Item:
File Description SizeFormat 
162997.pdfMemòria1.66 MBAdobe PDFView/Open

This item is licensed under a Creative Commons License Creative Commons