Latest / EHA Unplugged / AIDPATH – AI automated CAR T-cell manufacturing (with Dr Carmen Sanges)

Transcript

AIDPATH – AI automated CAR T-cell manufacturing (with Dr Carmen Sanges) · EHA Unplugged

• 0:03
  Greetings and welcome to EHA Unplugged, the official podcast
• 0:07
  channel of the European Hematology Association EHA.
• 0:11
  Hello, I am Isabella Livera, a medical writer at the European
• 0:15
  Hematology Association EHA. Welcome to this podcast on the
• 0:20
  advancement of an AI driven automated system for
• 0:23
  manufacturing car T cells. This this project has been
• 0:29
  founded by the EU and it's called 8 Path.
• 0:34
  Today I have the pleasure to be joined by Carmen Sanchez.
• 0:38
  Welcome. Could you please start by
• 0:41
  introducing yourself to our audience?
• 0:44
  So thank you, Isabella for giving me the opportunity.
• 0:48
  I'm Carmen Sanders. I'm the European Initiative
• 0:51
  Scientific Project lead for the Cellular Immunotherapy program
• 0:55
  in the University Clinic of Watzburg.
• 1:00
  I'm a medical biotechnologist by training with a PhD in medical
• 1:05
  biotechnology. 7 years of industry experience and then
• 1:11
  back to the university five years ago.
• 1:16
  Thank you for being with us today.
• 1:17
  To start things off, could you explain to us what is this
• 1:22
  Project 8 path? So 8 Path, it's a European
• 1:26
  funded project. We have been funded in the
• 1:30
  framework of Horizon 2020 for implementing an AI driven
• 1:35
  automated manufacturing platform that could overcome some of the
• 1:40
  challenges in the Karthi and in general in the TNP
• 1:44
  manufacturing. Together with 12 different
• 1:48
  partners that are experts with the very diverse background, we
• 1:55
  came together with the intention of building a physical platform
• 1:59
  that could help and support our aim of increasing access of the
• 2:05
  ATMP and European landscape. So the idea is to make the Carty
• 2:13
  or ATMP manufacturing process quicker, cheaper.
• 2:19
  Absolutely, because we are overcoming the manual processes
• 2:23
  allowing the the scale up of the manufacturing and we are also
• 2:30
  learning from the manufacturing process.
• 2:33
  So I would say that one of the biggest potential of this
• 2:36
  manufacturing platform, it's really the transparency of the
• 2:40
  data acquired during the manufacturing process.
• 2:44
  And for that, we basically rendered all the machines that
• 2:49
  are modular in this platform digitalised.
• 2:53
  And this thanks to the from over IPT partner, they developed a
• 2:58
  software that was able to communicate or to make
• 3:02
  communicating the different machines with each other, but
• 3:06
  also in correlation with a software that it's called the
• 3:09
  Corp software and it's also interfacing with the operator.
• 3:13
  And so I think that this is one of the major items together with
• 3:17
  the fact that this system is completely automated and it's
• 3:20
  completely closed. The manufacturing platform is
• 3:23
  basically consisting in 2 units and it's not only taking care of
• 3:28
  the manufacturing process, but it's also taking care of the QC.
• 3:32
  There is no disturbance for the cells whatsoever because we are
• 3:36
  basically operating all the all the necessary steps in one
• 3:41
  unique place. So which processes do you do in
• 3:46
  this manufacturer platform? Where do you start and where
• 3:49
  does it finish? It's end to end.
• 3:50
  So from the atherosis product until the car T product, so
• 3:56
  it's. To be injected into patient and
• 3:58
  it has been quality controlled and everything is ready correct
• 4:01
  it comes out. Great.
• 4:02
  Of course you need to sample on a certain point during the
• 4:06
  manufacturing and that's the other, the other beauty of the
• 4:11
  platform. You have the possibility to to
• 4:14
  have different ways to control what you are doing.
• 4:18
  The one that I find really interesting and exciting is
• 4:21
  looking at the cells during the expansion phase.
• 4:24
  The Aglaris bioreactor, the phaser prototype that was
• 4:28
  designed during this project and that allows basically a
• 4:32
  perfusion of the cells that it's very gentle.
• 4:36
  And it's also highly controlled because we embedded 6 different
• 4:44
  soft sensors that are able to capture metabolites and from
• 4:49
  from the cells. And given this inline sensors,
• 4:55
  we are also able to feed AI algorithms that are informing us
• 4:59
  on the feeding regime, that it's not fixed, but it's really based
• 5:05
  on the needs of the cells. And it's also informing us on
• 5:09
  the right harvest time. So the media is adapted.
• 5:13
  In real time. In real.
• 5:14
  To the needs of the cells. And we don't need to disturb the
• 5:18
  cells for taking samples because this is another thing that we
• 5:23
  also embedded at line controls. And there is an automated driven
• 5:30
  sampling system that is basically taking samples during
• 5:38
  during the process and it's allowing analysis on the
• 5:43
  platform itself. We included on the platform flow
• 5:46
  cytometers for example. And so you can check anything
• 5:50
  that you would like to to check in that in that specific moment.
• 5:54
  We also have the opportunity of automated sampling that is
• 5:59
  allowing you to make offline analysis.
• 6:05
  So the machine will do all these analysis automatically.
• 6:11
  Yes. So it's programmed to do it at
• 6:14
  central time points. It's all automated unless you
• 6:18
  want yourself to say, OK, now I prepare to have a sample to do
• 6:24
  some flow cytometry outside of the platform.
• 6:28
  So normally the platform would do everything, but you can also
• 6:31
  at some point decide we want to intervene.
• 6:34
  You can customize the schedule directly on the cope interface.
• 6:39
  So the interface is allowing also the scheduling of
• 6:43
  everything that it's on the manufacturing, including the
• 6:46
  sampling. So what else is AI doing?
• 6:54
  We have 5 different AI use cases for our manufacturing platform.
• 7:00
  The one that I told you is basically the one that it's
• 7:05
  utilizing the data from the expansion phase.
• 7:09
  So given the inline sensors in the in the facer and beyond
• 7:17
  that, we also have the possibility to deeply integrate
• 7:24
  the manufacturing platform into the hospital database.
• 7:28
  What does that mean? On top of the COP software, we
• 7:32
  also we are also backing the the manufacturing with a solid
• 7:36
  database infrastructure that is called Logic Suite that was
• 7:41
  developed by the the partner Arctic.
• 7:46
  In this database we have three different units.
• 7:50
  They're taking care of the data. The first one is taking care of
• 7:54
  the data recorded on the manufacturing platform and it's
• 7:58
  called Logic Cam. The the other one is taking care
• 8:02
  of the data recorded in the database of the hospital, so
• 8:06
  retrieving information from the patient history.
• 8:09
  And then we have also the logic analytics that it's putting
• 8:13
  together all the data for statistical analysis.
• 8:16
  Not only that, the logic analytics is also allowing to
• 8:21
  share research questions with other platforms.
• 8:25
  So in case of multiple manufacturing platforms, the
• 8:30
  database will collect the data in a Federated way or mocked way
• 8:36
  for all of them, allowing the operators to also have analytics
• 8:43
  and research questions on the data collected on multiple
• 8:47
  platforms. So the fact that this is
• 8:51
  connected to the hospital information must bring some data
• 8:58
  privacy issues. How do you deal with this?
• 9:01
  The data are SELDO anonymised. So in case of operators or P is
• 9:09
  they want to perform analysis on the data and they cannot
• 9:15
  actually go back to the single patients.
• 9:20
  And this is given by the fact that we are using a common data
• 9:22
  model for the collection of data.
• 9:28
  And another thing that we embedded actually in terms of
• 9:31
  AI, it's the scheduling, because this is also a very important
• 9:35
  problem for the resources of of the hospital.
• 9:39
  And so one of the AI use cases is really trying to manage the
• 9:44
  resources, the departments, the personnel and also the
• 9:49
  scheduling on of the process, the manufacturing processes on
• 9:53
  the platform. Now the pilot platform that it's
• 9:56
  sitting in in Bootsburg in the R&D environment is actually able
• 10:02
  capable of running just one thirty production.
• 10:08
  However, the idea is really to parallelise and to make possible
• 10:13
  multiple runs on the same platform.
• 10:16
  What you mean by scheduling is that, for example, you will
• 10:20
  schedule when you need the patient to come in and they.
• 10:22
  Need to be there for the manufacturing?
• 10:24
  When the manufacturing should start?
• 10:26
  When the FRS is scheduled to come, yes.
• 10:29
  Everything is done by the platform, yes.
• 10:31
  With AI help, OK with the AI algorithm?
• 10:35
  Absolutely. What do you think are the main
• 10:41
  advantages that you will get out of this platform?
• 10:45
  The first advantage is the proof of concept.
• 10:48
  So the FRS product, it's never the same and this is including
• 10:55
  the fact that if you have fixed manufacturing processes, you
• 11:00
  will always have different car T products that are not
• 11:05
  predictable, right. If you do not act on the
• 11:09
  manufacturing process, you cannot really predict the
• 11:15
  quality of the car T that you are releasing into the patient,
• 11:20
  whether this product can be really effective or persistent
• 11:26
  the way you wish. So knowing that, knowing that we
• 11:33
  could have more control on that part, I think that one of the
• 11:40
  biggest potential of this platform is really to introduce
• 11:43
  the adaptive manufacturing concept in the car T production.
• 11:50
  And I'm talking about car T production because the pilot
• 11:53
  platform it's been designed for the car T process and the mapped
• 11:58
  on the protocol of one of the clinical trials in the
• 12:01
  University Clinic of Woods Park that it's LION 1 clinical trial.
• 12:06
  We designed everything in the in that direction.
• 12:08
  But it it's really for proof of the proof in the concept in a
• 12:12
  first instance. It will be applied to other cell
• 12:16
  engineering cases and. It can absolutely, because the
• 12:20
  manufacturing platform is been designed such that really we we
• 12:26
  have a manufacturing module system.
• 12:29
  So the different machines can be plugged in and plugged out.
• 12:37
  But if you decide to modify your protocol, you can still exchange
• 12:43
  machines, exchange technology. And I think this is also another
• 12:47
  important aspect of of the platform because it's not a
• 12:52
  closed in highly integrated system, but it's something that
• 12:57
  is allowing you to adapt to the different protocols and to the
• 13:02
  different technology needs and updates.
• 13:06
  So it's fascinating that it's already a personalized
• 13:09
  treatment, but it will be even more personalized with the
• 13:12
  system because the process will be adapted to each patient,
• 13:17
  hopefully. That's the the direction we
• 13:20
  rolled out the the adaptive manufacturing concept with the
• 13:25
  feeding regime that we saved and and also with the harvest time.
• 13:30
  But we want to exploit the potential of these adaptive
• 13:34
  manufacturing even more. And for that reason, we will
• 13:40
  apply in the next future also metabolomic profiling of the
• 13:45
  Apharesis product and the Curti product in the expansion phase
• 13:52
  so that we can try to find ways to steer the metabolomic
• 14:00
  reprogramming already in the expansion phase.
• 14:04
  So. There are other projects of
• 14:08
  other automating systems that have been used or and also
• 14:13
  developed for this kind of technologies.
• 14:16
  How does your manufacturing system compared to this?
• 14:22
  Do you know? The HPEC platform is a real
• 14:26
  proof of concept for many research questions.
• 14:31
  Adaptive manufacturing, it's nothing that is out there yet
• 14:34
  because from a regulatory perspective, this is a journey.
• 14:37
  So we already, we already exchanged with the regulators in
• 14:42
  this regard and it was a quite informative and I have to say
• 14:49
  that there is nothing against this path, right, But it's still
• 14:55
  an unexplored field, I would say.
• 14:59
  So we're not trying to compete with commercialized products,
• 15:03
  but we are rather trying to really prove concepts that can
• 15:07
  be utilized in the in the next future to increase access, to
• 15:11
  increase scalability, to increase really production of
• 15:16
  the upcoming needs of the car T therapies.
• 15:23
  Automation is of course a big point close systems that are
• 15:29
  avoiding manual processes. So we are actually in line with
• 15:33
  what the the market is requesting right now or for
• 15:39
  example, yeah, the possibility of scheduling.
• 15:42
  I think also the manufacturing is, is also very, very nice
• 15:47
  feature. But I would say that really the
• 15:50
  possibility of stealing the the manufacturing process, it's the
• 15:54
  one thing that it's groundbreaking in this context.
• 15:59
  What really it's groundbreaking is that AI is allowing managing
• 16:05
  the conditions in an adaptive way.
• 16:08
  Totally. And the way that in the fact
• 16:10
  that you, you can have transparency on the data of the
• 16:14
  manufacturing data, I think it's really important because this is
• 16:17
  an emerging field. We are still learning from it
• 16:21
  and it's incredible in my opinion that manufacturing data
• 16:24
  are not acquired as much as they should.
• 16:30
  And so for that reason, I would really push for for that
• 16:36
  acquisition of the data in a systematic way and in a way that
• 16:43
  can be compared to it's absolutely the key, the key item
• 16:49
  for people working in the field, so.
• 16:53
  What are the next steps? Where do you see this going in
• 16:57
  the next five years? Well, the project will end this
• 17:03
  year December 2025. But so we are clear that we will
• 17:09
  need a follow up project for really exploiting all the
• 17:14
  potential of the 8 pet platform. We received the physical
• 17:18
  platform in Boots book and we are performing all the
• 17:22
  validation rounds. So we will also parallel in
• 17:27
  parallel a perform research, research on what we think, what
• 17:36
  should be the basis for the next metabolomic profiling and
• 17:43
  metabolomic reprogramming that we would like to further develop
• 17:48
  on the on the platform. We will also conduct in vivo
• 17:52
  experiments to already create the basis that will that will
• 18:01
  prove or support the hypothesis that cells generated or produced
• 18:08
  on the platform in this in this way are actually more effective
• 18:15
  as well. So you.
• 18:16
  Mean by in vivo in this case. Mouse.
• 18:18
  Mouse. Yes, you are not yet going to
• 18:21
  try these products in patient. So the the data that we
• 18:25
  generated already, of course, I cannot disclose much, but are
• 18:30
  encouraging to really continue in this direction.
• 18:36
  We are also lucky to to work with the Hospital Clinic
• 18:39
  Barcelona. You know that they are a great
• 18:44
  example, pioneering example for point of care manufacturing.
• 18:48
  They are pioneering for manufacturing of academic
• 18:54
  approved products. And so the, the experience that
• 18:59
  they have also in this regard, it's pivotal for us for really
• 19:04
  putting the the right step forward.
• 19:09
  So is there anything else that we haven't talked about that you
• 19:13
  would like to say about this AI driven system?
• 19:18
  I would reinforce the the concept that European projects
• 19:24
  can work that are a a potential really an enormous resource and
• 19:32
  enormous, enormous potential for the research, for the
• 19:35
  groundbreaking research, putting together different stakeholders
• 19:40
  and with different expertise and and background.
• 19:44
  It's an intense work, but it pays off.
• 19:51
  There is no single moment where you are thinking of your project
• 19:56
  like check boxes for deliverables or tasks that you
• 20:02
  need to accomplish because they were written in the.
• 20:05
  In the proposal, but you are rather really creating something
• 20:09
  meaningful for the field because you believe in it.
• 20:12
  And you, you're surrounded by people that if they have the
• 20:15
  courage to be in such a large environment and they're open to
• 20:20
  explore together, yeah, it means that they are, they are special.
• 20:26
  They are courageous. And we need more courage in
• 20:29
  research. Well, thank you very much Carmen
• 20:33
  Sanchez for this, for sharing this project with us and good
• 20:38
  luck in the next. Years.
• 20:39
  Thank you so much. Thank you to all of you for
• 20:42
  listening and stay tuned for the next episodes.