Siamese Neural Networks for One-Shot Drug Discovery



The application of deep neural networks in drug-discovery is mainly due to their enormous potential to significantly increase the predictive power when inferring the properties and activities of small-molecules. One major requirement to ensure the validity of the obtained neural networks models is the need for a large number of training examples per class. This invalidates the use of instances whose classes were not considered in the training phase or in data where the number of classes is high and oscillates dynamically. Unfortunately, this is a common scenario in drug-discovery, where the lead-optimization step is, inherently, a low-data problem, which makes it difficult to find potential analogous molecules with the desired therapeutic activity.
The main objective of this work is to optimize the discovery of drug analogues, with increased therapeutic activity, for the same pharmacological target, based on a reduced set of candidate drugs. We propose the use of a Siamese neural network architecture for one-shot classification, based on Convolutional Neural Networks (CNNs), that learns from a similarity score between two input molecules according to a given similarity function.
Using a one-shot learning strategy, we only need one instance per class for the network's training and a small amount of data and computational resources to build an accurate model. The preliminary results of this study showed that a one-shot learning strategy achieved consistent results given the low data available.


Computational Drug Discovery

Related Project

D4 - Deep Drug Discovery and Deployment (PI: Bernardete Ribeiro; co-PI: Joel P. Arrais)


Bioinformatics Open Days (BOD) 2020, February 2020

Cited by

No citations found