Self-Driving Synthesis of Protein Nanoparticles by Active Transfer-Learning-Assisted Autonomous Flow Platform

The ability to precisely engineer protein nanoparticles (P-NPs) with tailored size and low polydispersity index (PDI) unlocks their potential for a wide range of biomedical and material applications. However, conventional P-NP synthesis heavily relies on trial-and-error approaches, hindering efficient optimization. To address this challenge, we introduce a novel self-driving synthesis platform that leverages the power of active transfer learning and multitask Bayesian optimization (MTBO). This approach enables the intelligent transfer of knowledge from previous, even seemingly unrelated, synthesis tasks to accelerate the optimization of new P-NPs. In this study, comprehensive in-silico simulations demonstrate the significant acceleration achieved by pretrained MTBO models in optimizing the synthesis of four distinct P-NPs (BSA, Zein, Gelatin, and Hemoglobin). We further validate these findings by developing an autonomous flow microfluidic platform for algorithm-driven synthesis. Experimental case studies with BSA and Zein illustrate MTBO’s superiority over single-task BO in minimizing both size and PDI. Crucially, we introduce a mutual active learning platform where MTBO dynamically leverages data from the parallel synthesis of two P-NP systems. This strategy addresses the challenge of limited pre-existing data sets, further expanding MTBO’s practical applications in data-scarce scenarios. Our work highlights the transformative potential of active transfer-learning concepts in accelerating the synthesis and development of P-NPs and potentially a wide array of other advanced materials.