Using Machine Learning Models to Predict the Presence of Hereditary Cataracts and Pathogenicity from αB-Crystallin Mutations

Noah Civiletti, Samuel Cook, Maria Kinzie, and Elaine Liu

Abstract

Hereditary cataracts—where cataracts develop at a young age due to mutations in the αB-crystallin protein—are a relatively unknown, under-studied condition that affects tens of thousands of children per year. The process of detecting such cataracts in babies is also difficult and tedious, often leading to later treatment and increased risk of long-term vision problems. Furthermore, due to the lack of research, there is very little conclusive information surrounding detection of and reason for why specific gene mutations are pathogenic. This project dug deeper into such causes, ultimately seeing if a machine learning model could be created to predict the pathogenicity of an αB-Crystallin mutation from its biochemical characteristics and detect hereditary cataracts sooner. Using data derived by Biotechnology students at the UNC Wilmington Summer Ventures program, an Exploratory Data Analysis was performed to draw out notable characteristic variables, like the isoelectric point. Numerous covariate combinations were then tested with various cross validated machine learning classification models. Results suggested that the Support Vector Classification (SVC) model with “isoelectric point” and “evolutionary age as pdel” variables as covariates performed the best overall with an accuracy of 95%, considering feasibility for future data gathering and efficiency. The model had an accuracy that matched those of the runner-up models—K-nearest neighbor and adaptive boosting—and had faster runtimes. The most successful model was then used to create a predictor tool for observations to be classified in larger quantities.

Faculty Advisor: (Tracy) Cuixian Chen

Acknowledgment: Dr. Yishi Wang, Dr. Ying Wang, and the Summer Ventures biotechnology students

Watch the presentation (9:04)

Pose your questions to the researchers

Access Noah Civiletti, Samuel Cook, Maria Kinzie, and Elaine Liu's Discussion Forum

Supplemental Materials

SVSM_Poster_4_1_Maria Kinzie.pdf

contact banner icons.png Contact Info

Email: mkinz712@gmail.com

noahciv25@gmail.com