Project Leads: Kathryn Tabor and Jay Neitz, UW Medicine

Data Science Lead: Noah Benson

You pick a ripe, red strawberry outside in the bright sunshine, and shortly thereafter the same fruit still appears the same shade of red in the dimly-lit kitchen, even though the spectrum of light reaching the eye has changed dramatically. Color constancy—seeing a surface as the same color across large changes in illumination—is a central, unsolved problem in vision science. Our retina samples the colorful world with three cone types: long/red-, middle/green-, and short/blue-wavelength sensitive cones. Yet the dominant pathway exiting the retina has traditionally been viewed as carrying only two chromatic channels: red and green. Two channels cannot represent the full set of three-cone comparisons needed for stable color judgments across illumination changes. Recently we have mapped a novel connection in the retina that introduces blue-cone signals to the red/green circuit, creating a trichromatic channel in the retina. This raises the possibility that color-stabilizing computations begin in the retina, but the novel circuit’s function remains unknown. Testing this possibility is difficult because it hinges on human retinal wiring, which we cannot manipulate, nor evaluate independently of cortical color processing. We therefore are creating an in silico pipeline that generates controlled stimuli, runs competing two- and three-channel retinal circuit models, and trains readout models to make color judgments. It goes beyond “accuracy” to produce mechanistic explanations that identify which circuit ingredients are necessary for particular color perceptions, such as color constancy. The framework will serve as a tool for hypothesis testing and future studies of altered color vision in genetic disease.