Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

3D Convolutional Neural Networks for Computational Drug Discovery

Sunseri, Jocelyn (2021) 3D Convolutional Neural Networks for Computational Drug Discovery. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

Download (51MB) | Preview


This thesis describes aspects of the implementation and application of voxel-based con- volutional neural networks (CNNs) to problems in computational drug discovery. It opens by justifying the novelty of this approach by presenting a more mainstream approach to the common tasks of virtual screening and binding pose prediction, augmented with more sim- plistic machine learning methods, and demonstrating their suboptimal performance when applied prospectively. It then describes my contributions to our group’s development of voxel-based CNNs as we honed their implementation and training strategy, and reports our library that facilitates featurization and training using this approach. It continues with a prospective assessment of their performance, analogous to the first prospective evaluation, with the addition of a novel CNN-based pose sampling strategy. Next it makes a foray into model explanation, first in an oblique fashion, by examining the transferability of models to tasks that are distinct from but related to the tasks for which they were trained, and by a comparison with an approach based on exploiting dataset bias using other machine learning methods. Finally it describes the implementation of a more direct approach to model ex- planation, by using a trained network to perform optimization of inputs with respect to the network as a whole or individual nodes and analyzing the content of the result as well as its utility as a pseudo-pharmacophore.


Social Networking:
Share |


Item Type: University of Pittsburgh ETD
Status: Unpublished
CreatorsEmailPitt UsernameORCID
Sunseri, Jocelynjocelynsunseri@gmail.comjss97
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairKoes, Daviddkoes@pitt.edudkoes
Committee MemberHutchison, Geoffgeoff.hutchison@gmail.comgeoffh
Committee MemberChong, Lillianltchong@pitt.edultchong
Committee MemberKingsford,
Date: 6 January 2021
Date Type: Publication
Defense Date: 29 October 2020
Approval Date: 6 January 2021
Submission Date: 21 December 2020
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 177
Institution: University of Pittsburgh
Schools and Programs: School of Medicine > Computational Biology
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: Machine learning, virtual screening, cheminformatics, molecular docking, molecular modeling
Date Deposited: 06 Jan 2021 16:46
Last Modified: 06 Jan 2021 16:46


Monthly Views for the past 3 years

Plum Analytics

Actions (login required)

View Item View Item