Imagine trimming years off biologics development—and catching problematic formulations long before the first pipette is even picked up. That’s the promise of computational approaches in protein drug development, shaking the dusty traditions of trial-and-error and ushering in a smarter, more collaborative era.

For this episode, David Brühlmann welcomes Giuseppe Licari, Principal Scientist in Computational Structural Biology at Merck KGaA. A chemist by training, Giuseppe Licari pivoted from hands-on wet lab science to the predictive power of quantum mechanics and in silico modeling.

Today, he stands at the intersection of computation and CMC development, pioneering digital tools to streamline candidate screening, de-risk formulation, and ultimately bring therapies to patients faster.

Key Topics Discussed

• Predictive formulation tools reimagine experimentation as validation, allowing teams to foresee failures before entering the wet lab.
• “Good enough” stability reflects the realities of early clinical work, where safety, speed, and learning matter more than perfect shelf-life.
• Giuseppe’s journey—from theoretical chemistry to computational biology—illustrates how scientific curiosity can evolve into impactful in silico drug design.
• Computational scientists operate at the bridge between modeling and experimentation, turning predictions into actionable decisions through close collaboration.
• In silico developability assessments guide candidate prioritization by predicting safety, efficacy, and manufacturability risks early.
• Machine learning thrives on abundant, diverse data, while physics-based models remain critical for exploring new modalities where data are scarce.
• Modern computational modeling helps map how pH, salts, and excipients influence protein behavior, reducing reliance on trial-and-error formulation.

Episode Highlights

• Why maximizing shelf life isn’t always necessary in early development phases [02:56]
• The critical role of communication between computational and bench scientists [06:46]
• Core properties to assess for developability, including hydrophobicity, aggregation, charge, and immunogenicity [11:06]
• How accurate are in silico predictions, and where do they add the most value? [13:23]
• The limitations and strengths of machine learning and physics-based models in predicting protein behavior [15:19]
• The differences between developability, formulation development, and formulatability, and the value of early cross-functional collaboration [17:17]
• When to use platform formulations and when tailored approaches are needed for complex molecules [19:25]
• The advantages of using computational methods at any stage, especially for de-risking strategies [20:13]

In Their Words

The change in perspective is that we are now going from having several sequences in developability to having only a single sequence. So that’s the big change in discovery. We have several sequences, and now we need to apply methods to select only one. Then, in development, we have only that one selected sequence — we cannot change it anymore. So that is a very big change.

Historically, there has been a lot of work in the literature on mutating the protein to improve the characteristics of the API. But once the sequence is fixed, there is not so much in the literature on how we can support formulation development under that constraint.

Episode Transcript: From Developability to Formulation: How In Silico Methods Predict Stability Issues Before the Lab - Part 1

David Brühlmann [00:00:46]:
What if you could predict formulation failures before ever touching a pipette? Today we’re diving into the computational revolution transforming biologics development with Giuseppe Licari, who is a Principal Scientist in Computational Structural Biology at Merck KGaA.

From predicting aggregation hotspots to designing stable formulations in silico, Giuseppe reveals how computational approaches are slashing development timelines and catching problems that traditional methods miss.

Let’s explore how smart science is making formulation development faster, smarter, and more predictable.

Welcome Giuseppe — it’s great to have you on today.

Giuseppe Licari [00:02:42]:
Hi David, it’s my pleasure to be here with you, and thank you for the invitation to your podcast.

David Brühlmann [00:02:48]:
Giuseppe, share something you believe about bioprocess development that most people disagree with.

Giuseppe Licari [00:02:56]:
Well, in my field of drug product development, I believe we should set a “good enough” stability standard for our API to ensure we deliver the product safely and in a timely manner.

We don’t always need to maximize shelf-life stability — at least not for preclinical or Phase I studies. People might disagree and try to maximize shelf-life even early in the project, but I think that in Phase I we don’t need that.

Instead, we should aim to deliver the product as fast as possible, in a safe manner of course, to the patient and see if the project works.

David Brühlmann [00:03:46]:
Yeah, you’re making a great point — and I think it’s important to have a phase-appropriate approach, isn’t it?

Giuseppe Licari [00:03:52]:
Yes, because again, the problem is that we never know if a therapeutic concept will actually work, and we spend so much time and effort at the beginning of a project — and then the project may be stopped because there is no efficacy. So I think we need to target the right amount of effort according to the phase we’re in. And this is true for any function, for any step of the process. Of course, people have different views on this, but I think that as long as we deliver something safe for the patient, we are good.

David Brühlmann [00:04:26]:
I'm looking forward to diving further into today's topic — into developability and also formulatability. But before we do that, Giuseppe, let’s t alk about yourself, because your path from physical chemistry to computational structural biology is fascinating. So take us back to the beginning and tell us what sparked your interest — and what were some interesting pit stops along the way.

Giuseppe Licari [00:04:54]:
Yes, I think it started during my undergraduate studies, when I first encountered quantum mechanics and theoretical chemistry. I'm a chemist by education, and in those courses I discovered the fascinating capability of these techniques to predict molecular properties without performing any experiment.

From the computer alone, we could calculate something “out of the blue.” That was incredibly fascinating to me and sparked my interest in in silico and computational methods.

At the same time, I had a genuine interest in developing new drugs to help patients. So I tried to combine these two passions, and I became more and more interested in computer-aided drug discovery.

Of course, I also worked in the lab during my undergraduate studies and during my PhD, but over time I leaned more and more toward computational work.

A very important pit stop in my career was my three-year postdoc in the Theoretical and Computational Biophysics Group at the University of Illinois at Urbana–Champaign. I learned a lot there, gained extensive experience, and it opened up many perspectives for me. That was probably one of the most important parts of my career.

David Brühlmann [00:06:24]:
Can you paint us a picture? Because you're now at the intersection of computational biology and drug development, including formulation development. What does a typical day look like? Are you sitting in front of a computer all day doing modeling? Do you go into the lab? Is it a combination? What does that look like?

Giuseppe Licari [00:06:46]:
Yes — and importantly, what a computational chemist in a pharmaceutical company should not do is sit in front of the computer all day.

I truly believe it’s essential to constantly exchange with bench scientists, because we need to understand what is most valuable for them and where computational work can really make a difference.

So my daily work involves a lot of interaction with people in the lab, understanding their needs, and figuring out how computational approaches can support them.

Once we identify a need — for example, a specific screening or a particular question in a project — then I work on my side to carry out the in silico assessment. I provide my conclusions and recommendations, and then we discuss again and plan the corresponding lab activities together.

So it’s really a continuous exchange between the computational scientist and the lab scientist.

David Brühlmann [00:07:52]:
Let’s unpack this: developability, formulation development, in silico approaches. Starting from the very beginning — where do in silico approaches shine the brightest in drug development?

And when I say drug development, I include process development and the broader CMC landscape. You have seen many parts of biologics development, so where do you see the greatest benefit of these computational approaches?

Giuseppe Licari [00:08:23]:
First of all, in silico approaches are really vast, and there is a lot that can be done and applied in pharma. I’ll focus on the approaches that are most related to what I do. You mentioned this concept of developability. Maybe not everyone is familiar with it — it’s a relatively new way of thinking. We want to develop drugs that are safe, efficacious, and manufacturable, and the concept of developability was introduced a few years ago to help select a candidate with the highest overall developability profile.

From the experimental perspective, we can run many assays to understand how developable a drug is. However, we can also use in silico methods to screen properties of the API that can predict this developability profile.

So one major application is screening candidates in the final stages of discovery, when we might have, for example, 4 to 10 molecules. In silico methods can be very helpful in prioritizing the candidates — identifying the ones that might be more developable and more manufacturable.

Once the final candidate is selected and development officially starts, we can no longer change the sequence. But we can still apply several in silico approaches to help develop the best formulation. In this case, we don’t modify the sequence, but we can adjust what is around the API — the pH, the ionic strength, salts, surfactants, excipients.

So in silico methods can help filter out conditions that might not be favorable for your API.

David Brühlmann [00:10:25]:
And that’s such an important point — this concept of developability. For those who know me well, they know I’m really passionate about this topic because I strongly believe in starting CMC development early, already in discovery.

Doing this homework early and looking at the molecule’s properties ensures that it’s developable and, ultimately, manufacturable at larger scale.

Can you tell us what is typically evaluated in a developability assessment? What are the minimum protein characteristics you should analyze to make sure your molecule is developable?

Giuseppe Licari [00:11:06]:
Yes. There are several properties we can predict. For example, we can look at the hydrophobicity of the molecule and identify regions that are aggregation-prone; if necessary, we can mutate specific residues to remove these aggregation-prone motifs.

We can also predict the colloidal stability of the molecule — typically by looking at the charge distribution at different pH values, which gives us an idea of how stable the molecule might be in solution.

We can evaluate the chemical stability of the molecule, especially the residues in the CDRs — the complementarity-determining regions that interact with the antigen. These are crucial for antibody efficacy.

We can also assess immunogenicity, using several available computational techniques.

So yes, there is a wide set of properties we can predict, and these predictions can be very helpful in prioritizing the candidate. And you’re absolutely right that this must be done as early as possible — ideally with input from people in development.

This exchange between research and development is really critical, because development scientists can already provide insights related to the formulatability of the molecule. So it’s not only immunogenicity; it’s also whether the candidate can be formulated under the conditions required later in development.

David Brühlmann [00:12:41]:
Oh yes, absolutely — you’re speaking my language here. It’s absolutely crucial. And I’ve unfortunately seen projects where this wasn’t done early enough, and the consequences were severe.

So for anyone listening: start early. If you’re in R&D, communicate with process development and manufacturing colleagues early on to get their input.

Now, I’m curious — let’s take aggregation as an example. When you do these in silico predictions, how accurate are they? And how much wet-lab work is still needed to confirm them?

Giuseppe Licari [00:13:23]:
Sure. The predictions can be quite accurate, but of course no prediction is ever 100% accurate. It depends on the methods you use.

Some approaches are sequence-based, meaning you don’t need the structure of the antibody to predict aggregation. But you can also use the 3D structure, because residues that are far apart in sequence may be close in space and form hydrophobic patches — something you cannot detect from sequence alone. That provides additional insight.

A good way to improve accuracy is to combine information from different methods — integrating sequence-based, structure-based, and other computational models into a holistic assessment.

From my experience, hydrophobicity can generally be predicted quite accurately. However, it’s also important to note that experimentally, hydrophobicity is difficult to measure directly, because aggregation isn’t driven by hydrophobicity alone — electrostatics and other factors play a role.

So when comparing predictions against experimental results, we need to keep in mind that the experimental measurement is itself a composite of multiple contributions.

David Brühlmann [00:15:00]:
And I imagine that the more experiments you run and the more data you generate across different molecules, the better the predictions become — especially as you incorporate hybrid models or even machine-learning approaches to improve accuracy further.

Giuseppe Licari [00:15:19]:
Exactly. If you use machine-learning models, then you really need a significant amount of data. You can associate many properties of antibodies to those data sets, including electrostatic contributions, and this may improve your predictions. This is already being done in several methods.

I think the biggest challenge is actually finding the data — and finding data that is representative of all the possible APIs we might have in development. Nowadays we don’t only have standard monoclonal antibodies; we also have many multispecific formats, ADCs, and other new modalities.

The issue with machine learning is that, once you train your model on certain categories, the predictions may not extrapolate well to new modalities. That’s why I really like physics-based methods — because you can extrapolate. You don’t need experimental data to train the model; you rely on the underlying physics, and you can still generalize to new molecule formats.

David Brühlmann [00:16:36]:
Your work has now evolved from developability into formulation development and formulatability. We’ll talk about formulatability in a moment. I’m curious — how different is your work now, and your in silico approaches, when the goal is to develop a formulation? Having the right formulation is such an important part of CMC development.

So let’s start there. How different is this compared to developability? And then I want to move on to the next question: What are the specific approaches used to come up with a formulation that will work for your biologic?

Giuseppe Licari [00:17:17]:
The change in perspective is that in developability we start with several sequences, whereas in formulation development we work with only one sequence — the selected drug candidate. That’s the big shift. In discovery, we apply methods to select one molecule among many. Once we enter development, we can no longer change the sequence.

Historically, there has been a lot of work on modifying or mutating proteins to improve API properties. But once the sequence is fixed, there is much less guidance in the literature on how to support formulation development.

That’s the space you’re asking about — how to support formulation development using in silico methods. Now the idea is not to change the protein, but to change whatever is around it. The protein is fixed, but in a formulation it “feels” a specific environment — a given pH, buffer species, salts, excipients, surfactants. All of these may influence its behavior.

I am really convinced, and I have plenty of evidence, that simulations and computational approaches can help us understand what happens to a protein in a given environment. That’s the shift when moving from developability to formulation development in silico.

David Brühlmann [00:19:02]:
Earlier you mentioned a phase-appropriate approach. So how early should formulation development start? For example, in Phase I, should you use something “off-the-shelf,” like a platform formulation? I imagine this is easier for antibodies — but what about more complex molecules?

Giuseppe Licari [00:19:25]:
A platform approach can work for standard molecules — for example, for typical monoclonal antibodies. But when you have complex multispecific molecules, as we increasingly see in the clinic, it becomes more challenging. The platform formulation may or may not work.

In silico methods can be very helpful for de-risking your strategy and adjusting your planning. You can start with a broad platform and then use computational tools to filter out conditions that might be less favorable for your specific molecule.

Even for Phase I, you can use in silico approaches to fine-tune your strategy. The advantage is that you can apply computational methods at any time — you don’t need material, and they are relatively fast.

For Phase II, Phase III, or later stages, you can intensify experimental screening and rely more on computational support as needed. But at any phase, you can always go to in silico methods to gather useful information.

David Brühlmann [00:20:46]:
For those not familiar with formulation development, can you explain the difference between formulation development and formulatability? And when should each be performed? Or are they done together?

Giuseppe Licari [00:21:03]:
Formulatability is a relatively recent term, introduced in parallel with developability. It aims to evaluate whether a molecule can be easily formulated during development. So formulatability is assessed together with developability when screening candidates before development starts.

It gives you a forward-looking perspective: Is this molecule feasible to formulate under standard conditions? Or will it be challenging? That’s what formulatability tries to address.

Formulation development, on the other hand, is a work package executed during development — typically within the drug product development group. It is the process of identifying the best suitable formulation for a specific API. Any prior knowledge, including formulatability assessments, is extremely helpful for planning these experiments.

David Brühlmann [00:22:14]:
That wraps up Part One of our conversation with Giuseppe Licari. We’ve explored how computational methods are revolutionizing developability assessments and identifying formulation risks early.

In Part Two, we’ll dive deeper into excipient selection and real-world implementation strategies. If you found value in these insights, please leave a review on Apple Podcasts on your favorite platform.It helps other scientists like you discover these conversations. See you next time in Part Two.

All right, smart scientists — that’s all for today on the Smart Biotech Scientist podcast. Thank you for tuning in and joining us on your journey to bioprocess mastery. If you enjoyed this episode, please leave a review on Apple Podcasts or your favorite platform.

By doing so, we can empower more scientists like you. For additional bioprocessing tips, visit www.bruehlmann-consulting.com. Stay tuned for more inspiring biotech insights in the next episode. Until then, let’s continue to smarten up biotech.

Disclaimer: This transcript was generated with the assistance of artificial intelligence. While efforts have been made to ensure accuracy, it may contain errors, omissions, or misinterpretations. The text has been lightly edited and optimized for readability and flow. Please do not rely on it as a verbatim record.

Next Step

Book a free consultation to help you get started on any questions you may have about bioprocess development: https://bruehlmann-consulting.com/call

About Giuseppe Licari

Since 2022, Giuseppe Licari has been a Principal Scientist in Computational Structural Biology at Merck KGaA, where he leads efforts to build computational platforms for characterizing and screening biotherapeutic candidates. His work also explores how excipients influence protein stability, providing key insights that guide formulation development.

Before joining Merck, he contributed significantly to Boehringer Ingelheim by advancing in silico methods for developability assessment and predictive modeling of protein properties.

Giuseppe earned his PhD in Physical Chemistry from the University of Geneva in 2018 and later completed a postdoctoral fellowship with the Theoretical and Computational Biophysics Group at the University of Illinois at Urbana–Champaign, where he focused on simulating protein behavior at biological membranes.

Connect with Giuseppe Licari on LinkedIn.

He is also a biotech technology innovation coach, technology transfer leader, and host of the Smart Biotech Scientist podcast—the go-to podcast for biotech scientists who want to master biopharma CMC development and biomanufacturing.  

Hear It From The Horse’s Mouth

Want to listen to the full interview? Go to Smart Biotech Scientist Podcast. 

Want to hear more? Do visit the podcast page and check out other episodes. 
Do you wish to simplify your biologics drug development project? Contact Us