This audio article is from VisualFieldTest.com. Read the full article here: https://visualfieldtest.com/en/ai-found-21-new-genetic-clues-to-glaucoma-could-some-become-future-drug-targets Test your visual field online: https://visualfieldtest.com Support the show so new episodes keep coming: https://www.buzzsprout.com/2563091/support Excerpt: AI Found 21 New Genetic Clues to Glaucoma — Could Some Become Future Drug Targets? A study published in npj Genomic Medicine on September 19, 2026 used artificial intelligence to look for glaucoma-related patterns in optical coherence tomography scans, then tested those patterns against genetic data. The researchers reported 21 genetic loci not previously linked to glaucoma, and prioritized 11 genes that might help explain why glaucoma damages the eye. Five of those genes had not previously been reported in glaucoma research. () The work’s main advance is not that artificial intelligence diagnosed glaucoma better. It is that the researchers used measurable patterns of retinal damage—rather than a simple “glaucoma” or “no glaucoma” label—to search for genetic clues. That approach produced a richer biological picture, including a particularly promising signal involving RSPO2 and Wnt signaling in cells that help drain fluid from the eye. But these are genetic leads, not tested glaucoma treatments. The study did not show that changing any of the 11 genes prevents vision loss. The most exciting possibilities still require experiments to establish which genes are truly involved, whether changing them would help or harm, and whether a treatment can reach the right cells safely. What the study found—and what it did not The peer-reviewed article is “Deep-learning-derived glaucoma-related endophenotypes enable novel genome-wide genetic and functional discovery”. It reports 36 genome-wide significant loci in the European-ancestry analysis and 43 in a cross-ancestry analysis. The authors identified 21 loci as previously unreported in glaucoma and used additional analyses to prioritize 11 genes, five of them novel to glaucoma. () A locus is a region of the genome associated with a trait; it may contain several genes. A gene prioritized at a locus is a candidate for explaining the association, not necessarily the gene that causes it. The study therefore offers a set of hypotheses about glaucoma biology. It does not establish that the 21 loci—or any of the 11 genes—cause glaucoma, nor that a drug aimed at them would work. The methodological innovation, in patient-friendly language From a yes-or-no label to a detailed retinal “fingerprint” Many genetic studies classify each participant as having glaucoma or not having glaucoma. That can be a noisy comparison: some people may have undiagnosed disease, and people with glaucoma can have very different patterns and amounts of retinal damage. Instead, these researchers trained machine-learning models on eye scans from people with glaucoma. The models learned patterns in the ganglion cell complex and retinal nerve fiber layer—retinal structures that are affected as glaucoma damages the optic nerve. They then represented those patterns as quantitative traits, or measurable features of the scans. A useful analogy is the difference between asking whether a house is “damaged” and measuring exactly where its walls are cracked, how deep the cracks are, and how much the structure has shifted. The detailed measurements can reveal more about the process behind the damage. These scan-derived traits, which the researchers call endophenotypes, were then linked to genetic differences. They are glaucoma-informed measurements, not a replacement diagnosis or a measure of future vision loss. The study’s full-text manuscript describes two machine-learning model families, labelled AUTO and MOCO, used to produce traits from macular scans. () Reconstructing the research pipeline Train on glaucoma scans: The models were developed using 18,985 clinical optical coherence tomography scans from 8,323 glaucoma patients at Mass Eye and Ear. The earlier manuscript says most had open-angle glaucoma codes, though the training group was not restricted to chart-confirmed primary open-angle glaucoma cases. () Apply the models to UK Biobank: The final peer-reviewed publication reports 39,146 UK Biobank participants with usable genetic and scan-derived data. The ancestry groups were 35,739 European, 1,730 African, and 1,677 Asian participants. The large difference in group sizes means the European results had much greater statistical power. () Run genome-wide association studies: The researchers tested genetic variants against 21 scan-derived traits, separately by ancestry. A genome-wide association study looks across the genome for variants statistically associated with a measured trait. Split samples for discovery and replication: The manuscript describes a 75% discovery / 25% replication split within UK Biobank, followed by meta-analysis. This is useful internal checking, but it is not the same as replication in a fully independent cohort. () Combine results across ancestry groups: They conducted ancestry-specific analyses and then a cross-ancestry meta-analysis, finding 36 significant loci in the European analysis and 43 in the cross-ancestry analysis. Smaller African and Asian sample sizes limit how confidently those group-specific effects can be estimated. Prioritize possible genes and mechanisms: Follow-up analyses included gene-based testing, expression and splicing quantitative trait loci (eQTLs and sQTLs), Bayesian colocalization, Mendelian randomization, and single-cell expression enrichment in eye tissues. These analyses help narrow a locus to possible genes and cell types, but each has its own assumptions and limitations. () Which UK Biobank number should be used? The peer-reviewed publication’s figure is 39,146 participants. An earlier medRxiv abstract reported 47,908, but the manuscript’s detailed ancestry counts—35,739 + 1,730 + 1,677—sum to 39,146. The final publication’s participant number should be used for the peer-reviewed study. The public versions do not clearly explain the origin of the earlier headline figure, so it is safest to treat it as an earlier or inconsistent version figure rather than assume it represents the final analysis. () What “21 new genetic clues” means The authors call 21 loci novel to glaucoma because they had not previously been reported as associated with glaucoma or established glaucoma-related traits. “Novel” here does not mean the genes were newly discovered, that the DNA changes are unique to glaucoma, or that the genes’ roles in the body are now understood. The 11 prioritized genes were: EIF3E, LIN52, NPLOC4, PBLD, RSPO2, SLC25A16, TSPAN10, DYNC1I2, SLC25A12, ATOH7, and C14orf39. The paper identifies RSPO2, EIF3E, LIN52, DYNC1I2, and SLC25A12 as the five genes not previously reported in glaucoma. It also notes prior glaucoma-related or eye-trait evidence for some of the other genes, including ATOH7, C14orf39, NPLOC4, TSPAN10, PBLD, and SLC25A16. () The 11 prioritized genes: biology, eye-cell evidence, and drug prospects The cell evidence below refers to the study’s expression and enrichment analyses—not proof that a gene causes disease in that cell. The single-cell reference data came from non-diseased human eye tissues, and for several genes the article does not identify a particular eye-cell type as the key location. () For the full table, please open this article on visualfieldtest.com. Two shared regions make some gene assignments especially uncertain Several prioritized genes sit in the same genetic region. For example, DYNC1I2 and SLC25A12 share a locus in the study’s prioritization table; both may be plausible candidates, but the association does not establish which gene—or whether both—accounts for the signal. A similar caution applies to the chromosome 10 region involving ATOH7, PBLD, and nearby genes. These are reasons for follow-up experiments, not reasons to discard the findings. () The main biological themes Wnt signaling and aqueous outflow: the clearest treatment-oriented lead Aqueous humor is the clear fluid inside the eye. It normally drains through the trabecular meshwork. If drainage meets too much resistance, eye pressure can rise. Lowering pressure is an established way to slow glaucoma damage. RSPO2 is notable because it can strengthen Wnt signaling and the study found it in trabecular-meshwork and other anterior-segment fibroblasts, as well as optic-nerve-head fibroblasts. That puts a biologically plausible candidate in cells involved in tissue structure and outflow. Importantly, earlier laboratory work supports a role for canonical Wnt signaling in the trabecular meshwork and pressure regulation. In experimental models, blocking Wnt signaling raised pressure, and another study found that Wnt activation could reverse some abnormal features of trabecular-meshwork cells. Those results strengthen the general Wnt–outflow hypothesis—but they do not prove RSPO2 causes glaucoma or establish whether RSPO2 itself should be increased or decreased. () That direction matters. RSPO2 boosts Wnt signaling, and the prior eye studies suggest that too little canonical Wnt activity may be harmful in the trabecular meshwork. A simple plan to block RSPO2 could therefore be wrong. The best therapeutic idea, if the genetics is confirmed, may be to restore or tune local Wnt signaling, rather than turn it off across the body. Retinal ganglion-cell vulnerability, axonal transport, and mitochondria The study’s cross-ancestry gene-expression enrich Support the show