My name is Jia-En Chen, and I usually go by Danny. I am originally from Taiwan and am currently pursuing a PhD in Medical Physics and Biomedical Engineering at University College London. My research focuses on developing patient-specific, 3D-printed anatomical models for surgical simulation and training, particularly in neurosurgery.
My academic journey has always been interdisciplinary. I began with a background in public health before moving into automation and control engineering, medical 3D printing, and digital fabrication. In 2014, I founded the Medical Fablab at the National Defense Medical Center in Taiwan, where I worked with clinicians, engineers, and students to explore how fabrication technologies could address real healthcare challenges. I have also taught Python programming and medical 3D printing and contributed to clinical research, publications, patents, and educational programmes.
My connection with the HTBAA community began as an international student in 2016. The programme introduced me to synthetic biology and fundamentally changed how I understood the relationship between biology, engineering, and fabrication. Since 2022, I have continued contributing as a Global Teaching Assistant.
I am particularly interested in combining biological systems, advanced manufacturing, and human-centred design to create accessible tools for healthcare and education. I enjoy building collaborative communities where people from different disciplines can experiment, learn from one another, and transform ideas into meaningful real-world applications.
HTBAA Homework 1 Jia-En Danny Chen Department of Medical Physics and Biomedical Engineering, University College London
Project overview
I propose to biomanufacture standardised, living human neural networks on microelectrode arrays as an experimental foundation for biological brain chips.
My initial product would be an electrically responsive culture containing human neurons and supporting astrocytes. Each culture would occupy an individually addressable electrode well, allowing researchers to deliver electrical stimulation and record neural responses.
This idea connects my current research on 3D-printed anatomical phantoms for surgical training with my longer-term interest in living neural systems and biohybrid computing. My first manufacturing goal would be to produce 120 usable neural culture units for an on-site pilot study.
Starting with qualified, pre-differentiated cells, I estimate a manufacturing campaign of approximately seven weeks, including preparation, culture development and functional assessment. The estimates below are planning assumptions rather than experimental results.
1. What would I like to biomanufacture, and why?
The target product
I would like to manufacture a reproducible living neural network that can receive electrical stimulation and generate measurable responses.
Each unit would contain:
Human induced pluripotent stem cell derived neurons.
Supporting human astrocytes.
A cell-compatible culture surface.
A microelectrode array, or MEA, for electrical interfacing.
The biomanufactured product is the viable, connected neural network. Its electrodes, culture vessel and recording electronics would be manufactured separately.
For this proposal, one chip unit means one independent culture well with its own electrode field in a multiwell MEA plate. It is a functional manufacturing unit rather than a detachable semiconductor chip.
I would initially use a thin, predominantly two-dimensional culture directly on the electrodes. This provides a manageable starting point for investigating attachment, electrical coupling and manufacturing consistency. Spatially patterned, three-dimensional neural constructs would be a later development.
Connection to my previous research
My PhD research at UCL focuses on 3D-printed anatomical phantoms for surgical training, including skull-base and pituitary models. This work involves translating anatomical information into physical models, selecting materials, designing modular components and evaluating whether the resulting models are useful for their intended tasks.
These experiences have shaped my interest in the relationship between structure, material properties and function.
My current phantoms reproduce selected physical characteristics of tissue. I would now like to investigate how an engineered platform could incorporate living cells that develop, connect and respond over time.
The transferable aspects of my research include:
Designing controlled and reproducible physical environments.
Manufacturing modular components.
Understanding how geometry and materials affect performance.
I would apply these skills to culture fixtures, well geometry, electrode alignment and quality assessment.
However, the silicone, gel-wax and printing resins used in my existing models cannot be assumed suitable for neural culture. Materials exposed to cells would require separate compatibility testing. Neural culture and biological computing represent a future research direction, rather than work I have already completed.
Intended users and motivation
The initial users would be neuroscience and bioengineering researchers studying neural responses, stimulation strategies and biological information processing.
A useful first product would remain viable and produce repeatable, stimulus-dependent electrical responses. My longer-term ambition is to develop a biohybrid brain chip in which living neural networks interact with electronic hardware.
DishBrain and Brainoware provide research precedents for interfacing cultured neural systems with computational tasks [3,4]. They motivate this direction, but they do not establish that my proposed system will learn, reproduce human cognition or outperform conventional computers.
The first challenge is therefore to manufacture a consistent neural substrate on which these questions can be investigated.
2. What living system would I use?
Proposed manufacturing chassis
I would use a mammalian co-culture consisting of human iPSC-derived excitatory neurons and human iPSC-derived astrocytes.
Neurons would form the electrically active network, while astrocytes would support the culture and its development.
My initial formulation would contain:
Cell type
Viable cells seeded per unit
Intended role
Human iPSC-derived neurons
100,000
Electrical activity and neural network formation
Human iPSC-derived astrocytes
20,000
Support for the neural culture
Total
120,000
Living neural network
The 5:1 neuron-to-astrocyte ratio is informed by an existing MEA application study [5]. It is a starting formulation rather than an optimised specification for my device.
Manufacturing boundary
For the pilot, I would obtain qualified, cryopreserved, pre-differentiated cells from a suitable supplier or research collaborator.
The proposed manufacturing process begins with thawing qualified cells and ends with a functionally assessed neural culture.
This boundary excludes the supplier’s earlier iPSC expansion, differentiation and cell banking. A future process that includes these upstream stages would require additional time, resources and yield calculations.
Mature neurons would not be treated as an expanding biomass that can be multiplied through fermentation.
Inputs, biological processes and outputs
Stage
Description
Inputs
Banked neural cells, culture medium and supplements, a cell-adhesive substrate, coated MEA wells and controlled incubation
Biological processes
Cell recovery and attachment, neurite extension, formation of connections and development of electrophysiological activity
Output
A living neural network on an electrode interface, with associated manufacturing and functional assessment data
The cells would consume nutrients and produce metabolic waste as the culture develops. Regular medium exchange would therefore be required.
The output would remain a living culture that needs continued maintenance after manufacture. It would not be a shelf-stable product.
Why this living system?
A mammalian chassis is appropriate because the intended function depends on human neural cells and their interactions.
Microbial systems could potentially manufacture supporting biomolecules, and cell-free systems could produce selected proteins. However, neither alone would form the intended human neuronal network.
This makes mammalian culture a suitable choice despite its greater demands for time, environmental control and specialist infrastructure.
3. How would I use automation?
Liquid-handling automation
I would use an Opentrons Flex liquid-handling robot for repeatable medium preparation and partial medium exchange.
After validation, the robot could also assist with cell dispensing. This would depend on demonstrating acceptable cell recovery and dispensing consistency.
Opentrons describes automated media exchange and cell handling as supported application areas [7]. However, this neural workflow would still require custom labware definitions, pipetting validation and an appropriate aseptic environment.
The main automated tasks would be:
Task
Purpose
Medium preparation and dispensing
Improve consistency of volumes and reagent preparation
Partial medium exchange
Reduce repetitive handling and variation between wells
Cell dispensing after validation
Improve consistency of initial cell loading
Run logging
Record timing, plate positions and handling history
Data organisation
Connect manufacturing conditions with functional measurements
The robot would remove medium away from the cell layer and dispense fresh medium gently.
A technician would initially manage plate loading, incubation transfers and inspection. The liquid handler itself would not replace a tissue-culture incubator, an electrophysiology instrument or validation of aseptic handling.
Measurement and feedback
A separate MEA system with compatible stimulation capability would deliver electrical inputs and record responses. Microscopy would be used to monitor culture morphology and attachment.
I would compare automated maintenance against a manually maintained reference before adopting the process for routine production.
Important variables would include:
Cell settling during dispensing.
Aspiration height and liquid flow.
Evaporation.
Time outside the incubator.
Cell attachment after medium exchange.
The objective would be to make handling more consistent and to identify which manufacturing conditions produce reliable cultures.
Connection to the course readings
Petzold and Mukhopadhyay discuss automation, structured data and iterative exploration in microbial biomanufacturing [2]. I would adapt these engineering principles to neural cultures, while recognising that their review does not validate my specific mammalian process.
Asin-Garcia and colleagues emphasise the importance of shared research infrastructure and standardised workflows for translating biomanufacturing ideas into practice [1].
For this project, access to an appropriate cell-culture facility and electrophysiology expertise would be essential. I would not assume that any general-purpose cloud laboratory could maintain long-term neural cultures or perform the proposed measurements.
At larger scale, automation would need to coordinate incubation, plate transport, imaging and electrical recording, rather than only accelerate pipetting.
4. How much would I need to make, and how long would it take?
4a. Defining a meaningful scale
I would target 120 usable neural culture units for an on-site pilot study:
12 stimulation or measurement conditions × 10 units per condition = 120 usable units
This defines a practical manufacturing target rather than a statistically powered experimental design.
Units produced from the same cell lot and manufacturing batch would be technical replicates. Broader conclusions would require later experiments involving independent batches and donors.
Planning assumptions
Parameter
Planning value
Basis
Required usable units
120
Proposed pilot demand
Plate format
48 wells per plate
Selected manufacturing format
Number of plates
4
Calculated capacity requirement
Total culture capacity
192 wells
4 × 48
Reserved controls and sentinels
24 wells
Design assumption
Production candidates
168 wells
192 − 24
Production pass rate
80%
Unmeasured planning assumption
Cells seeded per well
120,000
Starting formulation informed by [5]
Recovery before seeding
85%
Unmeasured thaw and handling assumption
Working medium volume
0.300 mL per well
48-well precedent in [6]
Post-seeding culture window
42 days
Planning choice informed by [5,6]
The cell formulation and medium volume come from different application contexts. Their combination in the selected culture system would need experimental validation.
Number of starting units
At an assumed 80% pass rate:
Required production candidates = 120 ÷ 0.80 = 150 units
Adding 24 control and sentinel units gives:
150 + 24 = 174 occupied wells
Rounding up to complete 48-well plates gives:
4 plates × 48 wells = 192 wells
After reserving 24 wells, the expected usable output is:
(192 − 24) × 0.80 = 134.4 usable units
This is approximately 134 usable units, providing an expected margin of about 14 above the target.
The fractional value represents an average yield, not a physical output. This estimate does not guarantee delivery, particularly if failures affect an entire plate or cell lot.
Cell requirement
For budgeting, all 192 wells are treated as fully seeded, including controls and sacrificial sentinels. Any acellular blanks would slightly reduce actual cell consumption.
Neurons at seeding
192 units × 100,000 neurons per unit = 19.20 million viable neurons
Astrocytes at seeding
192 units × 20,000 astrocytes per unit = 3.84 million viable astrocytes
Total cells at seeding
19.20 million + 3.84 million = 23.04 million viable cells
Allowing for an assumed 85% recovery before seeding:
Required input = 23.04 million ÷ 0.85 = 27.11 million cells
This comprises approximately:
22.59 million neurons
4.52 million astrocytes
The 85% recovery factor represents losses before seeding. The 80% production pass rate represents a separate downstream loss. Neither should be interpreted as the final number of surviving cells per released unit.
Medium requirement
For budgeting, I assume:
An initial volume of 0.300 mL per well.
A 50% medium exchange, equivalent to 0.150 mL per well.
Exchanges on days 1 and 2, then every two days through day 42.
A total of 22 exchanges.
This pattern follows a manufacturer protocol precedent [6], but its suitability for the selected co-culture would need to be tested.
Fresh medium required per well:
0.300 mL + (22 × 0.150 mL) = 3.600 mL
For all 192 wells:
192 × 3.600 mL = 691.2 mL
Adding 20% for reservoir dead volume and handling:
691.2 mL × 1.20 = 829.44 mL
The estimated requirement is therefore approximately 0.83 L of fresh medium per campaign.
This excludes upstream cell production, washes, coatings, additional assays and maintenance beyond day 42.
4b. Manufacturing time
The baseline schedule assumes that qualified, pre-differentiated cell lots are already available.
Stage
Duration
Main activities
Preparation
7 days
Material preparation, incoming checks and automation readiness
Culture development
35 days after seeding
Recovery, attachment and neural network development
Release assessment
Final 7 days, through day 42
Repeated recordings, response assessment and batch review while culture continues
Total elapsed time = 7 + 35 + 7 = 49 days, approximately seven weeks
All four plates would be processed in parallel.
These are planning allowances rather than validated lead times. One application study reports quantifiable synchronous network firing from day 34 [5], while another protocol describes earlier network development for a different cell product [6].
This variation supports allowing six weeks after seeding rather than assuming a universally mature network within two weeks.
If acceptable function requires 56 days after seeding:
7 days of preparation + 56 days of culture = 63 days, approximately nine weeks
The estimate excludes procurement delays, initial process development and de novo differentiation from iPSCs. Establishing a reliable process for the first time could take substantially longer than manufacturing a batch once that process is established.
Automation throughput
I assume 45 minutes per 48-well plate for a validated partial medium exchange, including tip changes and cautious liquid handling.
This is an engineering allowance, not a manufacturer speed specification.
This excludes seeding and additional assays. Averaged across seven weeks, it corresponds to approximately 11 hours per week, although the workload would be concentrated on exchange days, including weekends.
For release assessment, an initial allowance would be:
This covers initial release recordings only. Complex learning experiments would require a separate instrument-time budget.
The overall manufacturing time would remain approximately seven weeks because neural development occurs over weeks, even when liquid handling is automated.
Scaling beyond the pilot
I would scale out by adding parallel culture units while maintaining their geometry.
With the same four-plate capacity and conservative seven-week campaigns, seven complete campaigns would fit within a year:
7 campaigns × 120 target usable units = 840 target usable units per year
This excludes initial process development and assumes timely turnaround and sufficient cell supply.
Larger production would require additional incubation capacity, coordinated handling and sufficient access to measurement equipment.
Sensitivity analysis
Scenario
Expected output or timing
Implication
80% pass rate with 168 candidates
134.4 usable units
Target met on average
60% pass rate with 168 candidates
100.8 usable units
Target missed
60% pass rate with 5 plates, reserving 6 wells per plate
210 × 0.60 = 126 usable units
Target met on average with 25% more wells and handling
56-day post-seeding culture window
63-day total campaign
Approximately nine weeks
These estimates indicate that functional yield and maturation time are critical manufacturing parameters to establish experimentally.
5. What do I not know yet?
Unknown 1 — Can I manufacture consistent functional networks?
The assumed 80% pass rate, 85% recovery and 42-day culture window have not been verified.
I would need to measure:
Cell attachment and survival.
Neuronal and astrocyte identity.
Electrical activity and stimulus responsiveness.
Variation between wells.
Reproducibility across independent manufacturing batches.
Provisional release criteria would include no detected contamination, acceptable cell identity in matched sentinels, at least 80% viability in matched sentinels, and reproducible stimulus-evoked responses above a predefined noise threshold in each released unit.
These are proposed development targets rather than established industry standards. Detailed thresholds and acceptable variability would need to be defined before evaluating production batches.
Unknown 2 — Will the materials and automated handling preserve the culture?
My previous phantom materials were selected for physical realism, not long-term neural compatibility.
I would initially use established MEA culture surfaces and keep printed fixtures outside the culture-contact region.
Any later custom material would be compared with a reference surface for:
Cell survival and attachment.
Culture stability.
Electrode coupling.
Effects on recorded neural activity.
Automated medium exchange would also be benchmarked against manual maintenance to quantify detachment, evaporation and contamination.
This would establish whether the proposed manufacturing hardware improves reproducibility.
Unknown 3 — Can the network support useful information processing?
Electrical activity alone is insufficient evidence of learning or computation.
Once manufacturing is reliable, I would investigate whether recorded neural responses can distinguish held-out stimulation patterns.
The evaluation would include:
Separate training and test data.
Shuffled-label controls.
Acellular controls.
A conventional computational baseline.
Analysis separating changes in the living network from learning performed only by an external software decoder.
Success would justify further investigation of patterned three-dimensional cultures and more demanding computational tasks.
Failure to demonstrate useful computation could still leave a valuable electrophysiology research model.
Human cell provenance, consent permissions and institutional review would be addressed before experimental work. Claims about consciousness or sentience are outside the proposed product specification.
References
Asin-Garcia, E., Fawcett, J. D., Batianis, C., & Martins dos Santos, V. A. P. (2025). A snapshot of biomanufacturing and the need for enabling research infrastructure. Trends in Biotechnology, 43(5), 1000–1014. https://doi.org/10.1016/j.tibtech.2024.10.014
Petzold, C. J., & Mukhopadhyay, A. (2025). From bench to biofactory: High-throughput technologies and automated workflows to accelerate biomanufacturing. Current Opinion in Biotechnology, 94, 103320. https://doi.org/10.1016/j.copbio.2025.103320
Kagan, B. J., et al. (2022). In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron, 110(23), 3952–3969.e8. https://doi.org/10.1016/j.neuron.2022.09.001
Charles River Laboratories, bit.bio, & Axion BioSystems. (n.d.). Functional phenotypic characterization of precision reprogrammed human iPSC-derived disease models for ALS using high-throughput microelectrode arrays. Application note. Read the application note
Axion BioSystems. (n.d.). CDI iCell human iPSC-derived neuron protocol. Manufacturer culture protocol. Read the culture protocol
HTBAA teaching team. (2026). Course overview and recitation transcript, and Homework 1 assignment brief. Course materials supplied with this assignment.
AI use disclosure
AI client: OpenAI ChatGPT, used through its Codex work environment, September 2026.
How AI was used: AI assisted with project framing, retrieval of course readings and technical sources, English drafting, engineering estimates and Markdown formatting.