Abstract
Purpose :
To use machine learning to identify gut microbiota associations with age-related macular degeneration.
Methods :
Macular OCT scans (Optovue iVue 100, Freemont, CA) taken between 2014-20120 from participants of the TwinsUK cohort were graded for signs of age-related macular degeneration according to the E3 OCT grading system.
Stool samples from TwinsUK participants were collected and DNA extracted and sequenced on an Illumina MiSeq platform. 16S sequences were demultiplexed in QIIME. Amplicon sequence variants (ASVs) were generated using the DADA2 package in R. Chimeras were removed and taxonomy assigned using SILVA 1.3.2. Samples with a sequencing depth of less than 10,000 reads were excluded.
Two different random forest analyses were conducted to select the most important microbiome features that predicted AMD status. The first ordered the microbiome variables based on the Gini classification index and the second used a frequency-corrected ranking approach.
Results :
A total of 404 stool samples from 321 AMD participants (mean age ±SD 67.6±7.8, 90% female) and 1,736 stool samples from 1,355 controls (mean age ±SD 61.3±7.6, 91% female) were included in the analysis. Our machine learning analyses identified three microbiota taxa that were likely significant predictors of AMD phenotypic status. Our results indicate that the phylum Firmicutes (RF permutational p-value=0.008) and additionally, two genera: Ruminococcus (RF permutational p-value=0.0005) and Prevotellaceae-NK3B31 are associated with AMD (RF permutational p-value=1x10-05).
Conclusions :
With the assistance of machine learning, this study suggests that an increased abundance of certain gut microbiome taxa may be associated with an increased probability of having AMD. Further validation is required from other cohorts and further studies are needed to explore any causal relationships between the gut microbiome and AMD.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.