Show simple item record

dc.contributor.authorFeng, Zhihao
dc.date.accessioned2021-12-20T20:34:25Z
dc.date.issued2021-08
dc.identifier.otherFeng_cornell_0058O_11213
dc.identifier.otherhttp://dissertations.umi.com/cornell:11213
dc.identifier.urihttps://hdl.handle.net/1813/110397
dc.description55 pages
dc.description.abstractResearchers identified amphiphilic copolymers as a class of promising material against biofilm formation in the recent decade. However, a guide for developing this type of copolymers remains unclear in many aspects (e.g., optimal pairs and compositions of hydrophobic and hydrophilic moieties). Upon the breakthrough of the experimental data collection with microarray, the journey of amphiphilic copolymers development could finally bring together artificial intelligence techniques. However, due to the biological condition differences throughout multiple data sources (e.g., we collected 2,420 experimental data entries for this study from seven different microarray experiments), the modeling generalization work has been a great challenge. In our approach with RDKit, our team constructed a synthetic feature dataset based on the experimental antibiofilm performances (quantified by logarithmic fluorescence intensities) of 2,420 polymers against \textit{Pseudomonas aeruginosa} PA01 (PA) and \textit{Staphylococcus aureus} 8325-4 (SA) suspensions. We then developed a data-preprocessing procedure, including a stratified dataset-split method and an autoencoder for feature dimensionality reduction. Furthermore, we optimized the hyperparameters of two radial-basis-function-kernelized support vector regression models in five-fold cross-validation with a random search algorithm for PA and SA, respectively. Upon satisfaction with the designed metrics in the final model evaluations, these two models demonstrated their strengths in capturing the correlations between the tested polymers and their experimental anti-biofilm performances with tolerant errors. In the end, we utilized these robust models to separately screen 2,745 unseen amphiphilic copolymers candidates and ranked their performances against PA and SA biofilm formation. After selecting the amphiphilic copolymer candidate of interest, we synthesized it with initiated chemical vapor deposition and recorded its anti-biofilm performance in the bacterial experiment. Based on these experimental results, we observed a certain-degree validation of these models for unseen polymers. This study provides a potential path to fast screen material candidates with AI, which has not been generalized by other computational models for amphiphilic copolymer high-throughput screening.
dc.language.isoen
dc.subjectAmphiphilic Polymers
dc.subjectBiofilm Formation
dc.subjectiCVD
dc.subjectMachine/Deep Learning
dc.subjectRDKit
dc.titleVirtual High-throughput Screening of Vapor-Deposited Amphiphilic Polymers for Biofilm Reduction with Machine/Deep learning
dc.typedissertation or thesis
dc.description.embargo2023-09-10
thesis.degree.disciplineChemical Engineering
thesis.degree.grantorCornell University
thesis.degree.levelMaster of Science
thesis.degree.nameM.S., Chemical Engineering
dc.contributor.chairVarner, Jeffrey D.
dc.contributor.committeeMemberYang, Rong
dcterms.licensehttps://hdl.handle.net/1813/59810
dc.identifier.doihttps://doi.org/10.7298/k0sn-d307


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record

Statistics