How historical AAV manufacturing data can uncover process optimization opportunities and help Process Development and MSAT focus their investigations
AAV manufacturing generates large amounts of data across manufacturing systems, process historians, batch records, in-process testing and QC laboratories.
But the data is often reviewed one batch or one system at a time. That makes it difficult to answer a bigger question:
Across all of our historical batches, what is actually associated with better manufacturing performance?
In this representative case study, Assurea used Manufacturing Intelligence and AI-based batch analysis to evaluate historical AAV manufacturing data, identify patterns associated with vector yield and product quality, and narrow the areas for further process investigation.
The analysis did not replace Process Development or MSAT. Assurea identified relationships in the historical data. The client then independently tested the findings through its own Design of Experiments (DoE), Process Development and MSAT scale-up activities.
Case Study at a Glance
| Type | Description |
| Manufacturer | Clinical-stage gene therapy manufacturer |
| Process | AAV viral vector produced using a suspension HEK293 transient transfection process |
| Historical data | 48 manufacturing batches |
| Data evaluated | 60+ manufacturing, process, in-process and analytical variables |
| Business question | What is driving batch-to-batch performance, and where are the strongest opportunities for AAV process optimization? |
| Assurea’s role | Historical batch analysis, pattern identification, process-to-quality correlation and post-DoE batch comparison |
| Client’s role | Process Development, DoE, MSAT scale-up, process decisions and representative-scale confirmation |
| Estimated duration | Approximately 5–7 months, including client-led experimentation and scale-up |
| Confirmation data | 4 representative-scale batches generated after Process Development and MSAT work |
| Representative outcome | ~15–25% higher vector genome yield and ~20–35% lower observed yield variability, while maintaining selected product-quality measures |
Case Study Note: This is a representative scenario based on realistic AAV manufacturing and process-development activities. The company, batch data and quantitative results are illustrative and are not actual Assurea client data.
How the AAV Manufacturing Intelligence Project Worked
The project followed a staged approach:
- Assurea analyzed the 48 historical AAV manufacturing batches.
- More than 60 manufacturing and analytical variables were evaluated across the batch population.
- Assurea identified five high-priority process relationships associated with batch performance.
- The client reviewed the findings and selected the areas it wanted to investigate.
- The client’s Process Development team independently designed and executed small-scale DoE studies.
- Promising conditions progressed through client-led Process Development and MSAT scale-up and confirmation.
- The client generated four representative-scale confirmation batches.
- Assurea compared the new batch data with the original historical population.
Estimated Timeline
| Stage | Owner | Estimated Time | Purpose |
| Historical batch analysis | Assurea | 4–6 weeks | Identify patterns, variability and potential performance drivers |
| Findings review | Client + Assurea | 1–2 weeks | Review and prioritize areas for investigation |
| Small-scale DoE | Client Process Development | 4–8 weeks | Experimentally test selected variables and relationships |
| Scale-up and confirmation | Client Process Development / MSAT | 3–6 weeks | Evaluate whether promising conditions translate as the process scales |
| Representative-scale runs | Client MSAT / Manufacturing | 4–8 weeks | Generate confirmation batches under the selected conditions |
| Post-DoE analysis | Assurea | 1–2 weeks | Compare the new batches with historical performance |
| Potential GMP implementation | Client | Outside case-study period | Evaluate manufacturing implementation through applicable client processes |
Actual timelines will vary based on the process, manufacturing schedule, analytical testing and availability of Process Development and MSAT resources.
The Challenge: 48 Batches, but No Connected View of Performance
The manufacturer was successfully producing AAV. The challenge was understanding why some batches performed better than others.
Each batch generated data across multiple areas.
Upstream and Transfection Data
- Cell density
- Cell viability
- DNA quantity
- Plasmid ratios
- Transfection reagent-to-DNA ratio
- Complexation conditions and timing
- pH
- Dissolved oxygen
- Temperature
- Agitation
- Culture duration
- Harvest timing
Downstream Process Data
- Harvest recovery
- Clarification
- Capture
- Purification and polishing
- UF/DF
- Final recovery
In-Process and QC/Analytical Data
- Vector genome titer
- Capsid titer
- Full/empty capsid measurements
- Potency
- Purity
- Host-cell DNA
- Residual plasmid DNA
- Other relevant impurity measurements
The manufacturer could review what happened within an individual batch.
What was harder to see was:
What consistently differentiates stronger-performing batches from weaker-performing batches across the entire manufacturing history?
That became the focus of the AAV batch analytics and Manufacturing Intelligence work.
Step 1: Building the Historical Batch Intelligence View
Assurea first brought the relevant historical manufacturing and analytical data into a structured view that could be analyzed consistently across batches.
This required aligning information such as:
- Batch identifiers
- Process stages
- Timestamps
- Process parameters
- In-process measurements
- Downstream recovery
- QC and analytical results
- Final batch performance
More than 60 manufacturing and analytical variables were evaluated across the 48 historical batches.
Instead of looking at each variable separately, the analysis examined relationships between process conditions, process behavior, downstream recovery and final analytical outcomes.
This created a connected view of AAV batch performance that could be analyzed across the full historical population.
Step 2: Identifying the Process Relationships Associated With Performance
The analysis compared stronger- and weaker-performing historical batches.
One important finding was that no single process parameter explained the differences in performance.
Instead, stronger relationships appeared when multiple factors were considered together, including areas such as:
- Cell density at transfection
- Transfection conditions
- Transfection reagent-to-DNA ratio
- Process behavior following transfection
- Timing-related variables
The analysis also evaluated these process relationships against product-quality results.
That was important because higher yield alone does not necessarily mean better batch performance.
Assurea looked at relationships involving:
- Vector genome yield
- Capsid performance
- Potency
- Selected impurity measurements
- Downstream recovery
From more than 60 variables, the analysis narrowed the data to five high-priority process relationships for the client’s scientists to investigate further.
The result was not a recommendation to immediately change a manufacturing setpoint.
It was a more focused question for Process Development:
Which of these relationships should we test experimentally?
Step 3: Finding Where Yield and Variability Were Being Lost
Final yield only tells part of the story.
Assurea also analyzed performance across individual manufacturing stages, including:
- Bioreactor production
- Harvest
- Clarification
- Capture
- Purification
- Polishing
- UF/DF
- Final vector
This helped distinguish between upstream productivity and downstream recovery.
For example, a batch could have strong upstream vector production but still finish with lower usable output because more material was lost during downstream processing.
Looking across the historical batch population helped identify whether performance differences were concentrated around production, recovery or a combination of manufacturing stages.
That gave the client more specific areas to investigate instead of simply trying to increase final yield.
Step 4: Moving the Findings Into Client-Led Process Development
Assurea presented the five priority relationships identified through the historical analysis.
The client’s Process Development team then determined which findings were scientifically meaningful and worth testing.
The client independently:
- Selected the relationships to investigate
- Designed the small-scale DoE
- Established experimental ranges and conditions
- Executed the experiments
- Reviewed the results
- Selected promising conditions for further confirmation
This was an important step because historical manufacturing analytics can identify correlations and patterns, but it does not by itself prove causation.
The client’s experiments were used to test whether the relationships identified in the historical data translated into actual process improvement.
Rather than beginning with dozens of possible variables, the Process Development team had a smaller, data-supported group of relationships to investigate.
Step 5: MSAT Scale-Up and Representative-Scale Confirmation
Promising conditions from the client’s Process Development work then progressed into MSAT scale-up and confirmation.
This step was necessary because results observed at small scale cannot automatically be assumed to behave the same way as the process scales.
The client’s Process Development and MSAT teams evaluated areas such as:
- Process consistency
- Mixing and agitation
- Transfection behavior
- Cell performance
- Harvest timing
- Downstream recovery
- Analytical quality results
Following this work, the client generated four representative-scale confirmation batches under the selected process conditions.
These were representative-scale confirmation batches, not four commercial GMP batches.
Step 6: Comparing the New Batches With the Historical Population
The resulting manufacturing and analytical data was provided back to Assurea.
Assurea compared the four representative-scale batches with the original 48-batch historical population to answer three questions:
- Did vector genome yield improve?
- Did batch-to-batch performance become more consistent?
- Were the improvements observed while maintaining the selected product-quality measures?
Representative Results
The four representative-scale confirmation batches showed:
- ~15–25% higher vector genome yield compared with the historical median
- ~20–35% lower observed batch-to-batch yield variability
- No observed deterioration in the selected potency, impurity and capsid-quality measures
The results provided an encouraging initial indication that the conditions selected through the client’s Process Development and MSAT work could improve both output and consistency.
Because the comparison included four new confirmation batches, additional manufacturing data would be needed to determine whether the improvement continues over time and at routine manufacturing scale.
Results at a Glance
- 48 historical AAV manufacturing batches analyzed
- 60+ manufacturing and analytical variables evaluated
- 5 high-priority process relationships identified
- Client-led DoE and MSAT scale-up completed
- 4 representative-scale confirmation batches analyzed
- ~15–25% higher vector genome yield
- ~20–35% lower observed yield variability
- Selected product-quality measures maintained
Why It Mattered
The value was not only the improvement in yield.
The manufacturer already had 48 batches of valuable process and quality data. The challenge was using that history to determine where scientists should investigate next.
The Manufacturing Intelligence analysis helped the client:
- Reduce the optimization search space by narrowing more than 60 variables to five high-priority relationships.
- Better understand batch-to-batch variability by comparing stronger and weaker historical batches.
- Separate upstream productivity from downstream recovery to better understand where performance differences were occurring.
- Connect manufacturing conditions with analytical results rather than optimizing yield in isolation.
- Focus Process Development experiments on areas supported by historical manufacturing data.
- Create a reusable historical batch population that future batches could be compared against.
Instead of manually investigating individual batches or testing a broad list of possible variables, the manufacturer could use its existing manufacturing history to make the next round of process investigation more targeted.
Beyond This AAV Optimization Study
The same connected manufacturing and analytical data can support additional batch-performance questions as more data becomes available.
Examples include:
- Comparing new batches with historically strong-performing batches
- Identifying where yield or recovery is being lost
- Analyzing relationships between process parameters and quality results
- Detecting changes in batch variability or process behavior
- Investigating relationships involving raw materials, equipment, deviations or cycle time
- Identifying earlier signals that may warrant technical investigation
This case study focused specifically on using AAV manufacturing analytics for batch performance and process optimization. Broader Manufacturing Intelligence and AI-based batch analysis use cases can be explored separately.
Frequently Asked Questions
What is Manufacturing Intelligence in biotech and pharma?
Manufacturing Intelligence brings together manufacturing, process and quality data so teams can analyze performance across batches rather than reviewing each batch or data source independently.
It can help manufacturers identify patterns, understand variability and find areas that warrant further process investigation.
How can AI and advanced analytics support AAV manufacturing?
AI and advanced analytics can evaluate historical AAV manufacturing data across process parameters, transfection conditions, downstream recovery and analytical results.
The goal is to identify patterns and relationships that help Process Development and MSAT determine where to investigate further.
Can AI identify the cause of low AAV yield?
Analytics can identify patterns and correlations associated with lower or higher yield, but correlation does not automatically establish causation.
Process Development and MSAT can use DoE and other scientific studies to experimentally test the relationships identified in historical data.
What data can be used for AAV batch analytics?
Depending on the manufacturing process and available systems, data can include:
- Electronic batch records
- MES data
- Process historian data
- Bioreactor parameters
- Transfection data
- In-process testing
- LIMS and QC results
- Downstream recovery
- Chromatography data
- Deviations
- Final analytical results
Does Manufacturing Intelligence replace Process Development or MSAT?
No. Manufacturing Intelligence helps Process Development and MSAT identify patterns and prioritize areas for investigation. The appropriate client teams remain responsible for scientific evaluation, experimentation, scale-up and manufacturing process decisions.
Turning Historical AAV Data Into Manufacturing Intelligence
A biotech manufacturer may already have years of valuable manufacturing data. The opportunity is to make that data easier to analyze across batches and connect it to the questions Manufacturing, Process Development, MSAT and Quality are trying to answer.
In this representative AAV manufacturing scenario, Assurea’s analysis turned 48 historical batches into a structured view of process performance, narrowed more than 60 variables to five priority relationships, and gave the client’s scientists a focused starting point for further investigation.
The client then tested those findings through its own Process Development and MSAT work, while Assurea used the resulting data to evaluate whether the expected improvement appeared in the new representative-scale batches.
That is the purpose of Manufacturing Intelligence and AI-based batch analysis: using the manufacturing data a company already has to uncover patterns, focus scientific investigation and build a stronger understanding of process performance.
Disclaimer: This case study presents a representative scenario to demonstrate how Assurea’s Manufacturing Intelligence and AI-based analysis services may be applied in AAV manufacturing. The organization, batch population, timelines, process-development activities and quantitative outcomes shown are illustrative and are not actual client data or guaranteed results. Actual approaches and outcomes will vary based on the product, process, development stage, manufacturing scale, available data and the client’s Process Development, MSAT, Quality and regulatory requirements.