Supporting Dataset for "Memorization Bias Impacts Modeling of Alternative Conformational States of Solute Carrier Membrane Proteins with Methods from Deep Learning"

Dataset Summary This Zenodo record provides complete modeling data and analysis outputs for the solute carrier (SLC) transporters described in the publication. The dataset contains 9 individual protein systems: (5OGE, 5I20, 6E9N, 6I1R, 6UKJ, ZnT8, SLC19A1, SLC19A2, SLC35F2) Each provided as a ZIP archive that includes subfolders for multiple modeling methods and derived analyses. Each protein folder contains results from: Modeller AlphaFold2 baseline predictions AF2-flipped-template AF2-notemplate Flipped-seq ESMFold AlphaFold3 AF3-flipped-template no MSA mmseq2 jackhammer Purpose and Methods The dataset supports an investigation into how memorization bias within deep-learning models influences the ability to model alternative conformational states of SLC membrane proteins.Predictions were generated using modified AlphaFold2 and ESMFold protocols with variations in template usage, MSA composition, and sequence orientation (“flipped-seq”) to sample diverse structural states. Software and Versions AlphaFold2: v2.3.0 (modified local implementation) AlphaFold3 Modeller: v10.3 ESMFold: Hugging Face/Colab (2024) Python: v3.8+ Supporting tools: Biopython, TM-score, PyMOL (for structure visualization and analysis) Abstract This study presents a computational pipeline for identifying alternative conformational states of solute carrier (SLC) transporters using deep learning-based structure prediction. We applied AlphaFold2 and ESMFold with custom input modifications (e.g., flipped sequences, altered templates) to explore conformational variability in pseudo-symmetric SLC proteins. This repository provides model outputs and supporting data for independent analysis and reuse. Supplemental Tables Supplemental Table 1: Supplemental Table 2: