The absorption spectra of human red and green visual pigments have peak wavelengths, λ(max), that differ by 31 nm, yet the opsins differ in only 15 amino acids. Mutagenesis studies have demonstrated that seven of the 15 amino acids determine the spectral shift. We trained neural networks to predict the λ(max) of any red/green chimeric protein. Seven mutants were excluded from the original training set. The trained networks were able to predict the λ(max) for the excluded mutants as an additional test, five new chimeric pigments were constructed and λ(max) determined. The neural networks correctly predicted the λ(max) of all five mutants. The use of neural networks is a novel approach to the problem of wavelength modulation in visual pigments.
Robinson, P. R., Griffith, K., Gross, J. M., & O’Neill, M. C. (1999). A back-propagation neural network predicts absorption maxima of chimeric human red/green visual pigments. Vision Research, 39(9), 1707–1712. https://doi.org/10.1016/S0042-6989(98)00289-2