Loading...

Molecular Property Prediction Using a Graph based Deep Learning Method

Shahcheraghi, Shamim | 2023

109 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56025 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Hossein Khalaj, Babak; Soleymani, Mahdieh
  7. Abstract:
  8. The goal of drug design is to identify new molecules with a set of desirable properties. The molecular search space is large, discrete, and unstructured, which results in a prolonged construction and testing process of new compounds and requires significant costs. Furthermore, there is a wide variety of appealing options to choose from. Recent advances in the field of machine learning have led to the emergence of generative models that, after training on real examples, can suggest suitable molecules with less time and cost. One of the stages that should be considered in the path of drug production is predicting the properties of the chemical molecule and its effect on the desired protein. By having a more accurate model to predict drug properties, the probability of drug development success increases. Using deep learning methods to predict drug properties is challenging because neural networks are data-hungry and need many examples of each feature to learn and perform optimally and with high accuracy. There are, however, a limited number of samples available in different datasets (1000-10000 chemical molecules). Additionally, most datasets tend to be biased towards a certain distribution within the chemical space, and in the available datasets, a limited number of molecules have properties in common. To solve problems whose data does not have enough and diverse examples, machine learning suggests methods. Meta-learning and self-supervised modules can be mentioned as examples of these methods. This project proposes a method for predicting the properties of chemical molecules that combines a meta-learning method with self-supervised and an attention module. It should be mentioned that chemical structures are displayed by graphs, and their analysis requires a neural network. It should be mentioned that chemical structures are displayed by graphs, and their analysis requires a neural network. In other words, by combining the mentioned modules including meta-learning, this system tries to optimize the graph neural network parameters in such a way that the obtained embeddings can be separated for chemical structures with similar characteristics
  9. Keywords:
  10. Molecular Properties Prediction ; Graph Neural Network ; Few-Shot Learning ; Metalearning ; Self-Supervised Learning

 Digital Object List