Machine Learning Prediction of Red Fluorescent Proteins (RFPs) - Data and Code Repository

This repository contains the datasets and computational notebooks supporting the manuscript "Machine Learning Models for Local Optimization of Red Fluorescent Protein Variants in a Low-Data Setting" for publication in the Journal of Chemical Information and Modeling. The work develops machine-learning models that predict properties of red fluorescent proteins (RFPs) — including brightness, emission wavelength, and Stokes shift — directly from protein sequence, and applies them to a single-mutation library of the mScarlet-i3 variant. Contents:- Curated RFP datasets: a sequence–attribute dataset, a multiple-sequence alignment, the mScarlet-i3 single-mutation library, and spatial/structural position labels.- Amino-acid descriptor sets (E-descriptor, T-scale, VHSE, Z-scale, and AAindex-derived principal components) and the optimized AAindex feature selections used by each model.- Jupyter notebooks for brightness and wavelength (emission and Stokes shift) prediction, covering AAindex/descriptor models, ensemble models, sequence-function comparison baselines (one-hot, ESM-2, UniRep/eUniRep), lineage-split cross-validation, and SHAP feature-importance analysis.- Underlying data for each manuscript figure (Figures 1–6). A full description of the repository structure, the software requirements, and instructions for reproducing the analyses is provided in the included README.