Native serotypes are suitable for projects with well-documented, cycle-limited and acceptable tissue distribution; engineered capsids are used to address receptor restriction, tissue barriers, immune evasion or off-target issues. Engineering strategies include rational design, random peptide insertion, DNA recombination, ancestry reconstruction, and machine learning design. Performance evaluation must be broken down into assembly, packaging, stabilization, receptor binding, tissue entry, cellular expression, and immune response.
Full screening workflow
|
Stage |
Core Movements |
Main phase-out risks |
|
1 Target product portrait |
Define species, cells, pathways, doses, off-targets and production indicators |
Blurred targets lead to drift in the screening direction |
|
2 Library design |
Control insertion sites, mutation depth, theoretical diversity and manufacturability |
A large number of unassembleable sequences |
|
3 Production and input library quality control |
Sequencing input library, titer, assembly and packaging bias |
Mistaking production advantages for organizational advantages |
|
4 In vivo selection |
Biological replicates, negative organs, multiple doses/time points |
Bottlenecks, random drift and off-target enrichment |
|
5 NGS and models |
Normalize by input library to differentiate between DNA arrival, RNA expression and function |
Label swapping, PCR bias and overfitting |
|
6 Monoclonal retest |
Independent preparation, blind comparison, histology and functional verification |
Pool filtering results cannot be reproduced |
|
7 Conversion Review |
Cross-strain/species, production scale-up, immunity and safety |
Early performance cannot be migrated across systems |
High-throughput sequencing and machine learning can expand the search space, but are not a substitute for rigorous input library calibration and independent validation. Reviews in the past three years further emphasized that incorporating assembly fitness, tissue performance and production attributes into multi-objective optimization is superior to ranking only by a single enrichment fold. [17-3 to 17-5]
References for this chapter
[17-1] Deverman BE, et al. Cre-dependent selection yields AAV variants for widespread adult brain gene transfer. Nat Biotechnol. 2016;34:204-209. [H]
[17-2] Ogden PJ, et al. Comprehensive AAV capsid fitness landscape enables machine-guided design. Science. 2019;366:1139-1143. DOI:10.1126/science.aaw2900. PMID:31780559. [H]
[17-3] Bryant DH, et al. Deep diversification of an AAV capsid protein by machine learning. Nat Biotechnol. 2021;39:691-696. DOI:10.1038/s41587-020-00793-4. PMID:33574611. [H]
[17-4] Guo J, et al. Computationally guided AAV engineering for enhanced gene delivery. Trends Biochem Sci. 2024;49:457-469. DOI:10.1016/j.tibs.2024.03.002. PMID:38531696. [N]
[17-5] Tan F, et al. Artificial intelligence-based approaches for AAV vector engineering. Adv Sci. 2025;12:e2411062. DOI:10.1002/advs.202411062. PMID:39932449. [N]