Abstract
Visual opsins are light-sensitive receptor proteins found in the eyes of fish, birds, reptiles, and mammals that enable animals to perceive color and adapt to their visual environments. These proteins absorb light most efficiently at their absorption maximum (λ_max), and mutations that shift or disrupt it can change color vision and contribute to inherited eye disorders, including color blindness. Accurately prediction λ_max and identifying the amino acids that control it will enable rational engineering of these proteins and provide insights into both vision disorders and the evolution of color vision. Traditional experimental techniques are time-consuming and labor-intensive, limiting the ability to study the large diversity of visual opsins. Computational models have therefore been developed to predict λ_max directly from protein sequence, but existing approaches often lack both predictive accuracy and biological interpretability.
To address this challenge, we developed a machine learning pipeline that predicts λ_max directly from a protein's amino acid sequence. We trained our models using amino acid residues located near the retinal chromophore (within 5 Å and 8 Å) and within the transmembrane core, regions that play key roles in strongly influencing λ_max. We benchmarked several machine learning algorithms, and the best performing XGBoost model, trained on residues within the membrane-spanning helices, achieved a coefficient of determination of 0.97 and average prediction error of 6.4 nm on the test set. We then applied SHAP-based feature importance analysis to identify the amino acid positions that contributed most strongly to predictions. Several highly influential residues, including F167, F213, F348, were identified near the retinal binding pocket, highlighting them as promising targets for future experimental studies of color tuning.
This framework provides an interpretable and accurate approach for predicting visual pigment properties and guiding future protein engineering. Beyond its potential applications in vision research, it offers insights into how animals evolved diverse visual systems, from species that perceive colors beyond human vision to deep-sea organisms adapted to extremely dim environments.