The Manuscript as Dataset: Biological Evidence and Computational Analysis in Medieval Palimpsests

Palimpsests offer rich textual layers for scholars of medieval manuscripts, but the physical signals of reuse often escape traditional visual or spectral detection. In this talk, I present a novel interdisciplinary approach that bridges molecular biology, computational genomics, and digital humanities to investigate whether mitochondrial DNA (mtDNA) preserved in parchment can serve as a material trace of palimpsesting processes and how machine learning might leverage such biological data to assist in computational classification. I begin by situating this work within broader digital humanities debates on material digital scholarship, where physical artifacts are reimagined as data sources through non-invasive sampling and computational analysis. Using non-destructive surface brushing, I collected DNA from parchment leaves and applied high-throughput sequencing and a robust bioinformatics pipeline to generate high-coverage mitochondrial genomes. Preliminary results show that palimpsested and single-use parchment yield nearly indistinguishable mitochondrial profiles, indicating that historical chemical and physical erasure techniques do not systematically degrade mtDNA. I then explore how supervised learning models, including neural networks, PCA-based analysis, and class imbalance strategies like SMOTE and sequence augmentation, perform on this biological dataset. While classifiers struggled to distinguish palimpsests from single-use samples based solely on mtDNA, the computational effort highlights the limits and potentials of biological data as a digital resource in manuscript studies. This project contributes to digital humanities by demonstrating how biological provenance data can enrich the interdisciplinary study of textual artifacts and by critically reflecting on the role of machine learning in interpreting complex cultural heritage data.