Data Institute Seminar: Deep learning applications in drug discovery and protein structure analysis

on the January 11, 2019

From 2:00 pm
Mehmet Serkan Apaydin (INRIA Sophia Antipolis) will give a talk on "Deep learning applications in drug discovery and protein structure analysis". The event will take place on January 11th from 2 pm in Pavillon Taillefer, room R32.
Recent developments in deep learning (DL) are mostly applied to problems in computer vision and natural language processing and are mostly developed by for-profit companies such as Google and Facebook. In particular, these two companies have developed autoencoders to encode words into so-called word vectors, that reflect the meaning of the words. With these word vectors, they aim to better understand what people write in their emails or Facebook posts. Recently, DL has started to be applied to problems in structural bioinformatics. In particular, multi-task neural networks and other DL frameworks have been used to predict molecular properties such as solubility or ligand-protein binding affinities, in low data settings. Furthermore, convolutional neural networks (CNN) have been used to compute fingerprints for small ligand molecules and these fingerprints have outperformed chem-informatics based ECFP fingerprints for molecular activity prediction. However, we are still in the early phases of the application of DL methodologies to this field, as the DL methodologies keep improving every year. In this presentation, I will talk about our ongoing work on using unsupervised learning approaches to obtain encodings for protein structures. With the resulting encodings, we plan to classify protein structures (fold classification). As in computer vision analysis of images with CNNs that result in a hierarchy of features, the features encoded with the DL architectures can represent protein structural motifs at various levels of details which we plan to study to explore the protein structural landscape. Similar encodings can also be used for representing protein binding sites and to look for similarities, resulting in the computational discovery of side effects and novel uses of existing drugs, which we plan to validate using existing databases.

Published on January 10, 2019

Practical informations


Pavillon Taillefer - room R32
5 avenue du Grand Sablon, La Tronche