Intelligent Affinity: Bayesian Optimization of “Full” AAV Purification by Convective Affinity Adsorbents

This project proposes an integrated platform to overcome current challenges with adeno‑associated virus (AAV) purification by combining convective affinity membranes (AvXcel®‑cellulose) with a Bayesian Optimization workflow (BEACON).

Industry Need

The gene therapy field faces major manufacturing bottlenecks, especially in downstream purification, where existing resin‑based affinity adsorbents do not enrich full capsids, suffer from slow flow rates, and require harsh elution conditions that reduce product quality. Additionally, current process‑development practices depend heavily on empirical, time‑intensive Design‑of‑Experiments methods that must be repeated for each new AAV serotype and provide poor interpretability and limited knowledge transfer. As clinical demand for AAV therapeutics grows, the industry urgently requires faster, more robust, and scalable solutions that can deliver high‑quality viral vectors with reproducible performance

Approach

The proposal delivers a dual innovation: (1) AvXcel® convective affinity membranes that enable high‑capacity binding at low residence time, significantly enrich full capsids at the capture step, and maintain caustic stability for extended reuse; and (2) BEACON, a machine‑learning‑driven optimization platform built on Gaussian Process models, transfer learning, and SHAP‑based interpretability. Together, these technologies provide a closed‑loop, data‑driven purification workflow that identifies optimal chromatography conditions rapidly, reduces experimental burden, and improves process performance. The platform will undergo academic and industrial testing before release as a validated NIIMBL resource

Value Statement/Outcomes

By integrating advanced affinity materials with machine‑learning optimization, the project will transform AAV purification into a rapid, efficient, and highly predictive process, reducing development timelines from months to weeks and lowering experimental effort by 30–50%. These advances will improve manufacturing throughput, reduce costs, and enhance product quality by consistently achieving high full‑capsid enrichment, strong impurity clearance, and robust scalability.