Using Machine Learning to Predict Response to Latanoprost
Certain single nucleotide polymorphisms (SNPs) increase the possibility of a positive response to latanoprost, whereas other SNPs up the chance of a negative response to the medication. Armed with that knowledge, a research team sought to find algorithms that would best predict latanoprost response. In an analysis involving 117 individuals, they concluded that predictive models could be useful. Their results were presented by Fernando Ussa-Herrera MD, a consultant ophthalmic surgeon at The James Cook University Hospital in Cleveland, UK, during the American Academy of Ophthalmology’s 2018 annual meeting in Chicago.
Participants had primary open-angle glaucoma and were treated with latanoprost >4 weeks. Investigators performed genotypic profiling using IPlex-Mass Array. They used data from 71 polymorphisms (SNPs) in 6 candidate genes (PTGFR and MMP-1,-2,-3,-9, and -17). Researchers calculated the predictive value for each SNP, and then used the 12 most significant SNPs. Machine learning techniques were used to develop 6 statistical predictive models. Among the results:
These 3 machine learning models showed the best predictive accuracy values:
- The linear kernel based on the support vector machine (SVM): 69, 23%
- Random forest: 66, 67%
- SVM cubic kernel model: 66, 67%
Ussa-Herrera F. Development of predictive models for latanoprost response in glaucomatous patients based on genetic variables. Talk presented at: AAO 2018 annual meeting; October, 26-30, 2018; Chicago.