PCSboost: A Multi-Model Machine Learning Framework for Key Fragments Selection of Channelrhodopsins Achieving Optogenetics

Abstract

Optogenetics combines optical and genetic methods to modulate light-controlled gene expression, protein localization, signal transduction and protein interactions to achieve precise control of specific neuronal activity, with the advantages of low tissue damage, high spatial and temporal resolution, and genetic specificity. It provides a cutting-edge approach to establishing a causal relationship between brain activity and behaviors associated with health and disease. Channelrhodopsin (ChR) functions as a photogenic activator for the control of neurons. As a result, ChR and its variants are more widely used in the realization of optogenetics. To enable effective optogenetics, we propose a novel multi-model machine learning framework, i.e., PCSboost, to accurately assist key fragments selection of ChRs segments that realize optogenetics from protein sequence structure and information dataset. We investigate the key regions of the ChR variant protein fragments that impact photocurrent properties of interest and automatically screen important fragments that realize optogenetics. To address the issue of the dataset containing a limited quantity of data but a high feature dimension, we employ principal component analysis (PCA) to reduce the dimensionality of the data and perform feature extraction, followed by the XGBoost model to classify the ChRs based on their kinetics, photocurrent and spectral properties. Simultaneously, we employ the SHAP interpretability analysis to perform an interpretability analysis of the ChR variant protein for pointwise, characteristic similarities to identify key regions of the protein fragment structure that contribute to the regulation of photocurrent intensity, photocurrent wavelength sensitivity and nonkinetic properties. Experimental findings demonstrate that our proposed PCSboost approach can speed up genetic and protein engineering investigations, simplify the screening of important protein fragment sections, and potentially be used to advance research in the areas of optogenetics, genetic engineering and protein engineering.