Listen on
Overview
Pharmaceutical manufacturing faces an inherent conflict: mass production for broad populations vs. individualized patient needs. The costs associated with fixed-dose, batch-based production are high, and true personalization remains a distant goal for most drugs. This challenge impacts patient outcomes, supply chain efficiency, and competitive advantage for biotech and pharma companies.
Jaidev Chakka, Principal Scientist at the University of Mississippi School of Pharmacy, is actively bridging this gap by applying 3D printing to create highly customizable pharmaceutical forms and biomedical devices. His work explores how data and AI can transform drug development and delivery, moving beyond traditional limitations.
In this episode, host Ross Katz speaks with Jaidev Chakka, who discusses the mechanics of 3D bioprinting for applications like bone regeneration and gene therapy. He explains the operational advantages of continuous manufacturing over batch processes, and how AI identifies critical printing parameters for quality and efficiency. The conversation projects a future where personalized ‘portapills’ are printed on demand, directly addressing individual patient needs through a continuously adapting data feedback loop.
Key Takeaways
Continuous 3D Printing Outperforms Batch for Pharma Manufacturing Quality
AI-driven studies demonstrate that continuous 3D printing processes yield superior quality compared to traditional batch methods in pharmaceutical manufacturing. This approach allows for real-time defect detection and parameter adjustment, drastically reducing waste and improving overall process control. The consistent environment of continuous manufacturing also minimizes variation sources present in batch setup and cleaning cycles, leading to higher product integrity.
Flow Percentage is the Critical AI-Identified Factor for Defect-Free 3D Printing
Through Design of Experiments (DOE) and AI analysis, researchers found that flow percentage — the amount of molten material pushed through the nozzle — is the most significant parameter influencing defect-free 3D printed tablets. This insight allows manufacturers to focus on precise flow control, refining print quality and potentially increasing print speed without compromising product integrity. It highlights how targeted data analysis can overturn initial assumptions about process variables.
AI-Driven ‘Portapill’ Represents the Future of Hyper-Personalized, On-Demand Medicine
The concept of the ‘portapill’ — an integrated system where real-time patient vitals inform AI models that then instruct a local 3D printer to create a precisely tailored drug dose — promises a fundamental shift in pharmaceutical delivery. This moves beyond fixed commercial doses to dynamic prescriptions adjusted daily, or even hourly, by a physician and instantly produced. Such a system requires reliable data feedback loops and secure, compliant manufacturing at the point of need.
Related: CorrDyn provides AI strategy and helps clients ensure data quality across their operations. Learn more about our work in biotech and life sciences or read our article on how biotech manufacturers gain data value.
Full Transcript
Jaidev Chakka: I think when it comes to 3D printing, having some advancements in the 3D printer space particularly for pharma manufacturing is very important.
Jason: Welcome to Data in Biotech, a podcast from CorrDyn where we explore how companies leverage data to drive innovation in life sciences. Every two weeks we sit down with an expert from the world of biotechnology to understand how they’re using data science to solve technical challenges, streamline operations, and further innovation in their business. Here we go.
Ross Katz: Jaidev Chakka, welcome to the Data in Biotech podcast.
Jaidev Chakka: Glad to be here.
Ross Katz: Awesome. Well just to get us started, would you mind giving us an introduction to your background and what brings you here to the APS PharmSci 360 conference?
Jaidev Chakka: Yes, I’m a principal scientist at University of Mississippi and I’m working with Dr. Mo Maniruzzaman in pharmaceutical engineering and 3D printing lab. This is my third time coming to this conference, which is a great venue to meet all the pharmaceutical scientists and researchers from academia, industry, everybody comes together and shares their experience, so it’s a great event here.
Ross Katz: Awesome. We’re going to spend a lot of time, you mentioned 3D printing, would love to hear a little bit about your research interests and some of the things that you’ve done in your work typically.
Jaidev Chakka: Sure. 3D printing is a unique technique where for me it is like printing something or creating something out of nothing. You have a filament which is a non-living matter and when you put it in a 3D printer it comes out into a beautiful shape and you can define the parameters of the shape, you can design it, you can control it. That fascinated me initially, and then when I got my hands on the 3D printer I thought, I have a background of biotechnology, so how can I apply that for biomedical applications. That’s when I started making scaffolds for bone regeneration, tissue engineering, 3D printing back in India when I was a postdoc at Indian Institute of Science in Dr. Kaushik Chatterjee’s lab. After that I got an opportunity to work at University of Iowa under Dr. Aliasger Salem, where I worked on gene therapy strategies. Taking a step ahead from a scaffold that just gives shape and structure, we were able to give a treatment in terms of functionality to that. That’s a great addition in terms of progression that I see how 3D printing as a technique is evolving. In the present lab I’m actually working on not only making 3D structure and giving it functionality, but we are able to have cells particularly spheroids and organoids — the scaffold is acting like a host giving space for the cells to grow inside them and this whole architecture even though it’s super complex acts like a kind of micro tissues which can be transplanted for regeneration purposes and can be used for preclinical testing, disease modeling, lots of opportunities.
Ross Katz: Interesting. I know that the applications of 3D printing to biotech are growing and in the process of coming into being. I would love to spend some time on some of the use cases that you just mentioned, understand them a little bit more deeply and then maybe we can talk a little bit about how data and machine learning and AI come into these applications of 3D printing. Does that sound alright?
Jaidev Chakka: Yeah.
Ross Katz: Scaffolding for bone regeneration — would you mind introducing us to the situation where a 3D printed scaffold for bone regeneration would be useful and how it compares to the way that we would typically do that sort of thing?
Jaidev Chakka: Sure. The gold standard for a bone defect or for the accident case or something is an autologous graft. A part of the femur bone taken and put in the ribs or some other place where it got broken can’t be chewed naturally and innately. Such defects are generally called critical defects. Anything bigger than 8mm in terms of defect volume in terms of diameter or length depending on the region, those are critical defects and they won’t reunite back, the broken two joints won’t grow and reunite back by themselves. You need to give some support. That’s where autologous grafts come. But it’s not like you’re taking some part of the body and putting it somewhere else — there is a loss of the component of bone in terms of mass overall. Now with MRI scans and CT scans you can actually create, regenerate the void that is there in the defect region and you can print exact graft that fits perfectly there. Having the material aspect of this 3D printed structure gives you the strength and shape, so the body, because it’s biocompatible, will immediately accept it and the cells migrate onto the scaffolds because they can identify it as something that they can rely on and they can grow, so that’s how the cells migrate and the whole regeneration process happens.
Ross Katz: Let me say that back to you to make sure that I’m understanding. Typically you take bone from another being, somewhere else, and you put it in the body, but anything longer than 8mm, the growth isn’t going to happen. So you’re 3D printing this scaffold that can be longer, it can be much larger, but it’s biocompatible, people’s bodies will accept it. The beauty of the scaffold is in the same way that we scaffold a building, you’re scaffolding the bone so that the cells can populate the gaps in the scaffold and grow into the fully fledged bone and then the scaffold, I guess, is consumed or falls away — is that what happens? Or does it stay there as a scaffold?
Jaidev Chakka: Just one correction.
Ross Katz: Sure.
Jaidev Chakka: In an autologous graft the bone won’t be taken from the other patient, it will be taken from the same patient to avoid immune rejection.
Ross Katz: Yes, okay.
Jaidev Chakka: When it comes to the scaffolding there is no chance of immune rejection because there is no component of that present, it’s just inactive material. Coming to your question, that’s a good question because the scaffold — if we take titanium kind of materials making into a 3D structure and putting it there, it stays there for eternity until the patient dies. If we take the biocompatible materials like polylactic acid or some other thermoplastic materials like PLG, PCL, PLGA, they actually degrade slowly in the body. Not at a faster rate but very slow. It’s like 1g of these materials takes two years to completely decompose in the body. That way slowly they just keep replacing — the cells replace this material with the actual calcium deposits that they grow over time.
Ross Katz: That’s amazing. You’re literally just giving the body time to rebuild the bone at the proper strength. Moving on to gene therapy strategies. Can you help me understand the intersection of 3D printing and gene therapy and how that works?
Jaidev Chakka: Sure. As I mentioned in the example, the cells need to migrate. Cells obviously want some stimulus. Like we are giving chocolate to the kids to listen to us. Something similar to that. When there is nothing — a void space, it’s not exhibiting any characteristic bone properties — the cells sometimes may take longer than usual to migrate onto these defect zones even though you give a 3D printed support. To accelerate that process for the cells to grow, come together and grow, we give some stimulating factors like bone morphogenic protein, also PDGF, VEGF and fibroblast growth factor. When we put these plasmids, even though one or two cells come, they express these plasmids giving the protein out from the cells. These act like chemokines where the cells attract towards these chemokines and come back. From one cell to 1000 cells, 1000 cells to a million cells. That way it accelerates the process.
Ross Katz: What I’m missing is the intersection of that with the 3D printing technology. What are you 3D printing in that case?
Jaidev Chakka: I understand your question. 3D printing, fused deposition modeling 3D printing that we do, involves high temperatures. 150 to 200 degrees centigrade. No biological component will be alive after coming into that environment. We’ll separate the things. First you print the scaffold and then you coat the scaffold with the biological gene therapy materials. How good is that coating? For that I used polydopamine. Polydopamine is a proteinaceous matrix that is exhibited by mussels. Mussels are generally considered a foul of the ocean because they stick to the ships, they keep degrading it, they reduce the flow of the ships, so cleaning them is a big hazard. If you analyze that mussels’ sticky nature that they have, what is the actual component that is allowing them to stick to the ship even though it’s speeding at 100 knots or what — I don’t know, I’m just assuming — that is the polydopamine component. Luckily this dopamine component we can polymerize from the individual dopamine hydrochloride units by manipulating the pH. We take the scaffold, put it in a dopamine hydrochloride solution and then change the pH. When you change the pH automatically the dopamine polymerizes, because it comes into contact with the scaffold, it has its ability to stick to any kind of material, it sticks to the scaffold and because of this nature the gene therapy plasmid complexes with the PEI also sticks to the surface. This dopamine coating is actually like a mediator, like a double-sided tape for both the scaffold and for the plasmids.
Ross Katz: Very interesting. You’re enhancing the scaffold to make it more functional inside of the body by coating it in this quote-unquote double-sided tape. My understanding there are at least a few companies here that are doing 3D printing for manufacturing applications. Can you explain from your experience, what are some of the most common or the most valuable applications of 3D printing to pharmaceutical manufacturing?
Jaidev Chakka: For pharmaceutical manufacturing the lab and the PI Dr. Mo Maniruzzaman actually initiated the 3D printing process for pharmacy applications. He envisioned a thing like in a home there will be a printer that continuously printed the formulations tailored to the patient. That’s the vision that was given to us. Coming to pharmaceutical manufacturing, 3D printing is a unique tool where you can put multiple drugs with multiple materials into a single tablet. In traditional manufacturing you have to evaluate every component — for just the drug to go into the body you need to give an excipient that will shield the drug and protect the drug, and during manufacturing you need to give some other components because when it’s compressed it has to come out, so you need to give some material. And for the compress in order to maintain strength you need to put another material and for the flavoring you need to put another material, if you want to give a color you need to put another material, all these things go on. Whereas in 3D printing it is just the material that shields the drug, biocompatible like HPMC, and the drug molecules. That’s all. The heat that the entire process undergoes simulates in a way, not exactly but the hot melt extrusion process where it amorphisizes the drug. This particular kind of technology is mainly suited for personalized medicines. For example, the commercial tablets are fixed doses like 100mg, 500mg. But for some patient’s condition they don’t need 500mg, they want more than 100mg. The optional let’s say 200mg or 250mg is not available. They have to break the tablet into half and take only half tablet. Here we can actually tailor that — this filament of this material of this concentration, if you give these dimensions of the tablet you will get this dose. Like that we can actually tailor the doses. For the manufacturing side it is challenging in a sense of speed and scale. Because tablets are currently manufactured in thousands of tablets per minute, whereas 3D printing one tablet will take three to four minutes for printing. If you want to scale that — say one printer prints one tablet at four minutes, if you want 1000 tablets, optimization-wise you need to have 1000 printers, all 1000 printers printing at a time, so you’ll have 1000 tablets in four minutes. Now if you want 10,000 tablets —
Ross Katz: Right.
Jaidev Chakka: That scaling aspect is what I feel currently is limiting in terms of even trying it out at a mass scale. But the technology is evolving in terms of 3D printing and pharma companies are also exploring other opportunities. One aspect that I observed is they are going into continuous manufacturing. With continuous manufacturing there is no batch. Materials are continuously coming into the reactor and they’re making it and getting it out. The report I read shows it saves a lot of cost for them in terms of operation side for the company. We did one study where we compared batch printing versus continuous printing, which one is best. In that study we used AI to see what parameters are actually helping. What we saw was continuous printing is better than batch printing. We are evaluating further why is that, but the two papers we published from the lab both show that continuous manufacturing is a step ahead from batch kind of manufacturing, but with 3D printers of course.
Ross Katz: You mean using 3D printers to print in batch versus using 3D printers to print continuously, one by one? Okay, that makes sense. Some of the things off the top of my head that might make continuous manufacturing better than batch are — when you manufacture in batch, you have to statistically sample from the batch in order to determine whether the batch is working or not, and then regulations force you to throw away the batch if you exceed a certain failure rate, whereas with continuous manufacturing you can understand on a one by one basis what are the predictors of failure and maybe throw away one dose, or you can detect that there’s failure happening as you’re manufacturing it and maybe even alter the process parameters to avoid that failure in the middle of manufacturing. Am I thinking about that right or what are some of the other contrasts that you would draw between batch manufacturing versus continuous manufacturing?
Jaidev Chakka: You’re absolutely right. Until now with batch, if one batch has gone bad, you can throw it. With continuous manufacturing when you do the direct comparison it seems like that may be a problem with the continuous manufacturing. But in continuous manufacturing when the process is going continuously there are no other factors coming in. For example in batch, once the batch is done, you have to clean it, you have to get it ready, and some error somewhere happens in terms of the cleaning not being good or something, which means the next batch is going bad. In continuous, once you set everything it’s going continuously, there are no breaks in between for any contamination or variation. As long as the material is coming up in a consistent manner, they were able to do that reaction and get the material out in a proper manner. Once it is regulated and it is deemed right quality, it goes like a good batch.
Ross Katz: That’s interesting. It’s not necessarily intuitive to me why batch machinery would be more likely to fail or less reliable than continuous process manufacturing machinery. Is it that in continuous manufacturing there’s specialization in the components that are happening along the lines? Because I would imagine that any individual process step could also fail. But maybe I’m thinking about it wrong — when you’re comparing 3D printer to 3D printer, am I making sense?
Jaidev Chakka: I understand your point. When it comes to 3D printing and traditional manufacturing, obviously the processes everything is different. Drawing a direct comparison may not be an ideal case, particularly in this scenario. But when it comes to 3D printing, having some advancements in the 3D printer space particularly for pharma manufacturing is very important. For example FabRx, they’re doing great in terms of making a GMP grade printer that can print doses. We are also in line to do something like what FabRx did, where we can have our own GMP printer manufactured and designed in USA.
Ross Katz: I want to return to 3D printing and the data that you collect from 3D printing so that we can understand the applications of ML and AI. Can you give us some insight into what are some of the data features that come off of a 3D printer that help you to understand what’s happening as you’re manufacturing a pharmaceutical?
Jaidev Chakka: Sure. We work with the fused deposition modeling method. There are different configurations of printers that work with that particular principle, like CoreXY, different configurations. The Cartesian printers, the configuration may be different, delta printers is different, but the principle is same where you take a hot plastic filament and you melt it and lay it into a shape or structure. That’s the core working principle of 3D printing. Layer by layer by layer. In this process when you observe, the printer speed is one parameter and the print temperature is one parameter and the flow percentage is one parameter and the material that you’re printing is another parameter that we took. When we printed we normalized the material because it is in direct correlation with the print temperature. The print speed, print temperature, infill percentage also took, infill percentage and flow percentage. What we did was we created combinations using DOE. Design of experiments, and we saw what are the least combinations that we can get to get an understanding of zero defects. We got 81 combinations. That’s a huge dataset that we need to make because we need to print, we need to convert those printed tablets into readable data for the algorithm. We used non-drug loaded placebo filament and made these 81 combinations and printed, and a graduate student named Shruti Chitnis was instrumental in spearheading this work. When we did that 81 combinations we took images and all the parameters of the tablets and when we put this into the adaptive design, these four parameters I mentioned for the printer, what value of each parameter can give us zero defects. The tablet should be solid, no defects, no pores, no holes, nothing. That’s the end goal because that’s the quality parameter we want for the final tablet. When we put all the data into the algorithm, it came out flow percentage is the key value, not print temperature, not print speed, not infill percentage, is the key value to get zero defects. In hindsight we know all these go hand in hand, these four parameters go hand in hand in order to give a good quality outcome from the 3D printer. But it is surprising to see the flow percentage is the value that is important for getting a good print. Because print temperature goes with the material, print speed goes with the printer, and in this case print speed has no major role to be frank. In our mind we thought when it is printing at high speed there may be a chance for more defects and at low speed it may be good, but here based on the predictions we got from the algorithm, an infill percentage with a little bit higher value at higher print speed is giving us better quality. In a way that saves time and we can get more tablets printed in less time without compromising the quality of the printed tablet.
Ross Katz: Can you unpack what flow percentage is and what the intuition is for why that would be the most important determinant of quality?
Jaidev Chakka: The flow percentage is generally considered for water. In the 3D printing space the molten material has flow properties but not the hot filament because it can’t flow. The molten material has flow properties and how much material the nozzle is pushing or flowing out through it is that percentage value. That percentage is calculated from the filament diameter and the nozzle diameter, and from these two values, lots of mathematical calculations calculate that percentage. It’s the percentage of how much it could potentially flow out. The filament is constant, the diameter is constant, material is constant, there is no variable there. But how much the material is coming pushed towards the heated nozzle — if the extruder is pushing more you get more material out of the nozzle. If it’s pushing less then it is less. Considering all these different components, the software gives the final percentage value. If you alter the percentage value it tells the extruder to push less and obviously the material coming out of the nozzle is also less. We played with that material to see that value. For example in manufacturing scales, if there are some defects happening, then obviously you can guide it like okay push a little bit more in order to fill the space so there won’t be any defect.
Ross Katz: That makes a lot of sense. If I’m thinking about that right, then hypothetically you could have the 3D printer adapt in this case to the flow percentage to manage the flow percentage to stay relatively constant, and the implication there would be that you would minimize defects by keeping the flow rate exactly the same. Are there any other machine learning and AI applications to 3D printing that you think are really relevant or exciting?
Jaidev Chakka: Two things that I feel are exciting. We used AI for understanding what parameters are relevant to get defect-free printing on a 3D printer. Can we apply the same design in order to understand what are the key parameters that influence the release of the drug? It’s not like we don’t know, but we have a dataset with hundreds of papers — not only hundreds, to be frank — in pharmaceutical research with 3D printing in the last decade you find so many papers. There was a group in UK, Dr. Abdul Basit, their group actually published a paper where they used the literature as a dataset and did an AI algorithm and predicted certain combinations in order to see the outcome in terms of the release profile. That was done, but taking a step ahead can we combine in terms of manufacturing and also in terms of release profile. We are in the second leg now, to see if we give these predicted parameters on the printer, if we give different materials and the drugs, what are the predictions that the algorithm can give in order to get a defect-free tablet that gives a tailored dose and a release profile that is suited for the patient. That is one interesting application that we want to move ahead into. The other one is we are also working on spheroids, cell spheroids, mesenchymal stem cells for the bone regeneration I mentioned, I used mesenchymal stem cells for differentiation towards osteoblasts. Are the stem cells as spheroids different or the same? One understanding even though not published, what we have is after making HMS spheroids they’re way different than the regular HMSCs in terms of retaining their stem cell nature. Cells, particularly primary cells like mesenchymal stem cells, the passages — they grow and then you split them into multiple flasks for them to grow more, so each passage they can grow only four to five passages. After that they quit, they differentiate into any other cell that they like, mostly to osteoblasts. But when you preserve them when you make spheroids of HMSCs and you culture them a week and then grow them back in the petri dishes, the passage numbers were higher. That’s something interesting to me — maybe in the body they’re not a monolayer, maybe when you put them in their native space in any kind of thing, they are a bit more active and able to give more passages, able to grow more generations. There we want to have AI in terms of optimizing that process because optimization is the key step in order to get a quality outcome. We can work into different parameters, the growth time, the growth media, growth conditions, morphology of the cell through fluorescence microscopy imaging or confocal, and we can get an understanding of how these parameters are actually influencing the shape. If it is a spheroid then it’s the sphere growth, the volume, the number of cells that we’re getting there, all those things, and extrapolating the same to the organoid. That’s a very interesting space because organoids are tiny organs that have both the cell phenotype of the organ and also the functionality of the organ. In a way you are working with the organs at a micron scale in the petri dishes — not exactly micron scale but in the petri dishes. If we can harvest the intelligence — if it is a neuronal organoid they can respond to the stimuli. When it’s an electrical signal there is always intelligence there, that was proved by computers. These functional units can act as individual processing entities, where we can create intelligent systems around that and optimize those systems, its behavior and also how much input, how much output and what is that output, how to interpret that output. I feel like AI/ML can act like a GPU for the CPU of these organoids on top of this application.
Ross Katz: First of all I want to zoom out and make sure that everybody’s tracking. The organoids are, like you mentioned, miniature versions of the organs on micron scale, and then you mentioned one version of that is you can have an organoid like neurons that are firing, that have an electrical signal, and that electrical signal can be understood as computation. One of the applications of AI and ML to that computation is this idea that you can use a machine learning model to train the organoid to produce computational spikes. What might help is, what are some of the applications that you would see of having organoids that could do that kind of computation?
Jaidev Chakka: One particular example from Dr. Thomas Hartung’s group was evaluating the toxicity of the developmental neurons. From my understanding, if I read the paper right and understood right, generally for those kind of studies you need to go with animal models. You need to understand during development how these brain’s neurons work, if you put the toxicity what happens. But when they have these organoids and with the AI adaptation where they used the electrodes that can capture the signal and these electrodes can also transfer the signal back to the organoids, these innovations are important for facilitating these kind of studies. They were able to identify how these toxins are affecting the ability of these neurons to fire up and fire down. It avoids the use of animals per se for the animal care.
Ross Katz: That’s really interesting and hypothetically a methodology for a new form of assay. If you can train the neuronal cells to fire in certain situations and not fire in other situations, and you can using AI and ML control the parameters on which that firing happens, then you might be able to, similar to the way that fluorescence is used to detect all sorts of things in biology, also use this method to detect other things besides toxicity inside of these organoids. Am I thinking about that right?
Jaidev Chakka: Your understanding is right. In our whole body everything everywhere is there. What I mean is if we take heart, cardiomyocytes are the major component. But you have blood vessels, you have neurons, you have other components like epithelial cells and everything. Even though when we say we are working on heart research, if you take just cardiomyocytes, okay that’s good, but it’s not complete. With organoids we can either differentiate into other cells also, or with IPSCs, induced pluripotent stem cells, we can differentiate into cardiomyocytes and you can try it out because it preserves that innate nature of the cardiomyocytes rather than culturing them from few cells from the heart. Those are more relevant. Increasing the complexity-wise, what we understood was from the organoids you can actually measure the toxicity based on these firing patterns — can we extend this to other organs? Having a brain organoid and a heart organoid and a liver organoid, when you give a dose to the liver organoid, what happens to these ones? These interesting thoughts give some enthusiasm to look into those directions. It is possible with the kind of advancements we’re seeing in microfluidic ways and also the electrodes that I mentioned. Without those electrodes we can’t do that kind of studies. There is a company that was actually doing that, Cortical Labs, they’re making digital brain systems. As time goes by, these technologies are going more mature and going much more advanced. That gives at least a little bit of hope for somebody like me to think in that direction to see maybe one day I can do that.
Ross Katz: That makes sense. You’ve got this multielectrode array, you’ve got a bunch of electrodes that can both receive signal and send electrical signals, and then you can connect that to a computer with a model on it, and the goal in a certain case would be to train this organoid to detect something that’s happening biologically inside of the organoid system that you just described. Can you help me understand how would that training process work to the extent that it’s known? I recognize that this is at the frontier of the field, but how do you teach the neuronal organoid to send and receive the signals when it’s detecting for example toxicity?
Jaidev Chakka: It’s already proved with various publications and by researchers. For example one publication from Nature, I believe, from Dr. Brett Kagan where they use brain organoids to play ping-pong. They were able to train that, making them play the video game. It may sound silly if you hear it first time, but when you see the research it’s fascinating. The kind of things they were actually doing — even for that reason Cortical Labs were actually making digital brain systems for preclinical evaluation kind of thing where we can put different materials and see how the brain reacts. Instead of going with the actual animal models or the brain of a living being. Training these organoids obviously involves electrical signals. How much signal you’re giving to the cell and how much is coming out, how much the cell is giving you out. When it comes to the memory aspect of these neuronal cells, on top of my head I don’t have all the information — for example if you take the brain, the cerebrum, cerebellum, medulla, medulla oblongata is the seating center for the entire brain where it stores the information and processes it. If we consider the same for the organoid it won’t be at that scale but it is able to still process it. When we give the training for the cell in order to store that signal for a fraction of time, that’s their memory. When you give that command in a digitized form at a particular voltage to that array, to that organoid in the whole array, it stores only one bit of information. You have 100 organoids storing 100 bits of information, and when you put that — I don’t know how that machine learning code in that electrical signal format — these 100 organoids some fire some not, and the outcome that comes from there is an output signal for it to operate.
Ross Katz: I think I might need to read up more on the subject. The idea of gradient descent or backpropagation in the context of organoid learning is something that’s very fascinating to me but also hard for me to understand based on where I am in terms of my biological understanding.
Jaidev Chakka: You’re good. I’m also learning so much.
Ross Katz: You’ve described your lab in our previous call as taking baby steps into AI. For your lab and for the scientists that are doing work in your lab, what’s been the steepest learning curve in terms of implementing AI and ML for your team?
Jaidev Chakka: Having the right people that were able to understand what we are trying to achieve. For example, the paper that we worked on for pharmaceutical dosage forms with AI, we worked with Dr. Samrat Choudhury from school of engineering at University of Mississippi, and before that we hired some data scientists, a couple of them, but their understanding is from the side of data, not exactly what we are trying to get. If we give a bunch of information to them they are ready to process it, but they’re not able to ask us, what is the data that I want in order to do this application for the pharmacy side? This professor has this understanding and he was able to ask us all the right questions and we were able to give him all the data that he needed. That’s how we were able to get that done. The steepest curve is being a traditional biotechnologist or pharmacist — these are unexplored and new venues for us even to think in these directions. That may be the steepest learning curve for us, to understand who wants what and how to communicate to them to get the outcome.
Ross Katz: You have a particular experiment you’re trying to run or biological phenomenon that you’re trying to measure, and this person’s coming in saying, well you need the person who can ask the right questions about what can be measured, about how it’s being measured, about what’s being predicted and how to connect the dots between those things. Am I thinking about that right?
Jaidev Chakka: You’re right.
Ross Katz: As we come toward the end of our conversation, I’m interested, if we fast forward two to five years, what breakthroughs do you hope to see in AI-enabled pharmaceutical manufacturing or organoid development or personalized medicine that you think have the most transformational potential based on the research that you’re doing?
Jaidev Chakka: We may see something in a direction called maybe instant manufacturing. Like instant coffee — everything is ready, you just have to put all the ingredients there, your coffee is ready. Something like that. The filaments that I mentioned previously, when we load drug into them, they protect the drug, keep them stable. When you have these filaments like cassettes at your disposal in a particular machine — for example if you need aspirin during the middle of the night, simple drug but still you don’t have it, you need to get it. Having a cassette of that, you can just press the button and there you have it. For example some critical doses that you need to take in defined very complex doses, you can print them back in the same printer having all these filaments ready. We are actually devising in a different direction, having a printer that can print multiple doses at a time. Doses in the sense not individual tablets of different doses, but also the same tablet with different doses of the drug or different drugs themselves. That we call Portapill. There we need some software advancements of course with the integration of AI. That’s going to be a beautiful thing to see because AI continues with the feedback mechanism as it is learning continuously, and it can adapt to the changes that are happening in a patient’s health. It can actually manage better in my opinion.
Ross Katz: I want to unpack that a little bit to make sure that I’m understanding correctly. We touched on personalized medicine earlier, I want to really open it up now. When a person takes a manufactured dose of a pharmaceutical, let’s say they’re consuming it orally, it’s decomposing, it’s dissolving in their stomach, it’s getting into their bloodstream. People are metabolizing it differently based on where they are in their lives, based on what’s going on with them healthwise. My understanding is there’s this opportunity to personalize in a variety of different dimensions. There’s personalizing for you and your biology, this is the dose that’s meant for you, but then also the dose that you should be taking or the nature of the pharmaceutical that you should be taking might vary from day to day and week to week, so there’s this opportunity to integrate the Portapill or personalized manufacturing with the status of the patient at any given point in time. Am I thinking about that right, or what are the other vectors of personalization that you would think about?
Jaidev Chakka: You’re absolutely right in this scenario. As the patient needs change, the dose changes, and as the doctor prescribes them to okay today take only 5mg less than what it was yesterday. I think that’s possible with the Portapill. It’s an integrated system where the vitals are continuously monitored by the physician and any drop in them or any change in them, the physician will have the information of that and can actually assign — he can instantly say okay take this much dose of this and send it directly to the printer. The printer will print and the patient is ready to take the dose.
Ross Katz: From a commercial perspective, the pharmacy doesn’t need to be open, it can just be a vending machine with a 3D printer in it, you just go in and this is who I am, I need my dose.
Jaidev Chakka: Isn’t that futuristic? At one dose it feels right.
Ross Katz: But there’s just so much work to be done on a variety of dimensions. There’s the assays of being able to understand what’s happening with the patient without having to take huge vials of blood or use very expensive mass spec equipment to understand what’s happening, and then there’s inputting that into the feedback loop of the manufacturing systems.
Jaidev Chakka: You’re absolutely right. Traditional manufacturing has thought of a lot of things — it has standard process and the USP, FDA giving the guidance documents and all those things, and USP giving us the monographs on different drugs and other things. Over the years all these regulating agencies and particular bodies have done a fantastic job in order to have a normalization in terms of the use and also the application evaluation quality everything. Now with 3D printing if I come and say hey you don’t need all that, the doctor will tell you — that’s not how it is. We have to get these processes and evaluations compatible with the regulating agency guidances, and only then can we move ahead because ultimately everybody wants the patient to have a good quality of life. Some may think okay I don’t need this, I do this way — no, that’s not the case, because we have to ensure the safety and efficacy of the drug and the safety of the patient in every step that we are doing. Even though we are dreaming big or having some bold ideas here, we still have to do each and every step that is required to ensure the patient safety and also the compliance in terms of drug handling, excipient handling, manufacturing standards, everything.
Ross Katz: That makes a lot of sense. Well Jaidev, I really appreciate you coming on the podcast today. Where can listeners go to learn more about your work?
Jaidev Chakka: They can go to Google Scholar — a lot of Jaidev, if they type that they can get all my publications there, and also LinkedIn the same.
Ross Katz: Awesome. Well Jaidev, thank you so much for coming on the podcast today, I really appreciate it.
Jaidev Chakka: Thank you very much for having me, appreciate it, thank you.
Jason: And that’s it for this episode of Data in Biotech. If you enjoyed the episode, please subscribe, rate, or leave a review in your podcast platform of choice. See you next time.
