Improving AI-based Synthesizability Scores for Next-Generation Protein Degradation Drugs

Tingting Mo and Jia Xin Zhu — Chalmers University of Technology and University of Gothenburg, 2026. Examiner Rocío Mercado.

Can you tell whether a proposed PROTAC is actually makeable without running a full retrosynthesis search every time? The thesis builds a systematic framework for PROTAC synthesizability assessment: 27,099 PROTACs curated and labelled, decomposed with PROTAC-Splitter and scored with AiZynthFinder, then distilled into Random Forest, XGBoost and MLP surrogates. Component-level and whole-molecule synthesizability agree 57% of the time, with linkers identified as the dominant source of synthetic difficulty. The best classifier reaches 0.958 ROC-AUC; the best regressor reaches an R² of 0.497, rising to 0.661 once noisy labels are filtered.

See Supervision for the full list of MSc theses I’ve supervised.