Back to search

Cornell University

COMPUTER AIDED RETROSYNTHESIS FOR METABOLIC PATHWAYS

Abstract

dc:description.abstract

My work centers around predicting the outcome of biochemical reactions using deep learning models. Specifically, we are interested in studying how a prior hypothesis of the reactions transformation can be effectively incorporated in the model framework. Biochemical retro-biosynthesis has met with limited success as there is no guarantee of discovery for every reaction. Over the years several retro- biosynthesis frameworks have been developed which have explored different aspects of reaction structure. Traditional biochemical synthesis methods rely on human intervention leading to an infinite number of possible outcomes. However, attempting to confirm them all experimentally is impractical. This is where retro- biosynthesis comes in, that is, the framework helps generate the most promising targets. The process of ranking improves the chances of discovery. Herein, we present how knowledge of the pathways can be used to restrict the number of possible candidates by ranking them based on likelihood of stability, which could be validated experimentally later. We used a deep learning network to predict the stability of a reaction taking the reaction as the input. The model was tested on a set of 100 distinct reactions, from 18 pathways outside of the training dataset, and it provides recommendations for the optimal choice of pathway. Candidates are ranked based on four types of models: reaction level, atomic level, molecular level, and spectator model. We created a preference list for models based on the type of metabolic pathway and enzyme usage. The utility of the models is illustrated through several examples from different reaction types. We also highlight the generalization of model performance to non-natural pathways. Through the use of extended connectivity fingerprints and descriptors’ properties, we have shown how the model’s performance can widely vary depending upon which parts of the molecule’s structure are features for the model. We have tried to highlight the significance of the spectator and actively participating parts of a reaction for determining the reaction outcome.

Degree

thesis:*
Name thesis:degree_name
M.S., Chemical Engineering
Level thesis:degree_level
Master of Science
Discipline thesis:degree_discipline
Chemical Engineering
Grantor
Cornell University
Year dc:date.issued
2020

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Wani, Aasim Ayaz
Chair dc:contributor.chair
  • Varner, Jeffrey
Committee members dc:contributor.committeemember
  • Bindel, David
  • Li, Sijin

Rights

dc:rights
Statement dc:rights
  • Attribution 4.0 International
Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
16938311
Dc Identifier Other
ProQuest Submission ID: 11131
ProQuest Publication ID: 28261329

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Wani, Aasim Ayaz. COMPUTER AIDED RETROSYNTHESIS FOR METABOLIC PATHWAYS. Master of Science thesis, Cornell University, 2020. https://hdl.handle.net/1813/117352