---
title: "Quality in, quality out: How biophysical data powers AI-driven formulation optimization for next-gen biologics"
description: "Learn how combined rigorous biophysical characterization with a multi-objective Bayesian framework helps identify optimal biologics formulations in just 33 experiments."
canonical: "/quality-in-quality-out-how-biophysical-data-powers-ai-driven-formulation-optimization-for-next-gen-biologics-blog"
published: 2026-10-22
---

# Quality in, quality out: How biophysical data powers AI-driven formulation optimization for next-gen biologics

[https://nanotempertech.com/](https://nanotempertech.com/) Solutions Choose Solutions Application Products Consumables Application Products Consumables Biologics [Be confident you’re choosing the right biologics candidates.](/biologics/) Products [Prometheus](https://nanotempertech.com/prometheus-panta/) [Learn more](/biologics/) Targeted Protein Degradation [Advance your TPD and induced proximity research.](/tpd/) Products [Dianthus](https://nanotempertech.com/dianthus/) [Dianthus uHTS](https://nanotempertech.com/dianthus-uhts/) [Learn more](/tpd/) Membrane Proteins [Characterize membrane proteins in complex environments with confidence.](/membrane-protein/) Products [Monolith](https://nanotempertech.com/monolith/) [Prometheus](https://nanotempertech.com/prometheus-panta/) [Learn more](/membrane-protein/) [Protein Stability Prometheus Panta One experiment. Multiple insights.](/prometheus-panta/) [Affinity screening Dianthus Tough targets. Confident decisions.](/dianthus/) [High-throughput screening Dianthus uHTS Screen faster. Discover more.](/dianthus-uhts/) [Affinity measurement Monolith Minimal sample. Maximum insight.](/monolith/) [Protein stability Andromeda Crude lysates. Early decisions.](/andromeda/) [Consumables Capillaries and plates designed for high reproducibility. Go to shop](https://shop.nanotempertech.com/) [Reagents Labeling kits and reagents for straightforward assay set up. Go to shop](https://shop.nanotempertech.com/)Technologies [nanoDSF](https://nanotempertech.com/nanodsf/)[Dynamic Light Scattering (DLS)](https://nanotempertech.com/dls-7/)[Static Light Scattering (SLS)](https://nanotempertech.com/sls-6/)[Backreflection](https://nanotempertech.com/backreflection/)[Spectral Shift](https://nanotempertech.com/spectral-shift/)[MicroScale Thermophoresis (MST)](https://nanotempertech.com/mst/)[Temperature Related Intensity Change (TRIC)](https://nanotempertech.com/tric/)[Resources](https://nanotempertech.com/resources/)[Support](https://nanotempertech.com/support/)[About](https://nanotempertech.com/about/)[Shop](https://shop.nanotempertech.com/) EN English汉语 EN English汉语 Search Search{results\_count} resultsDisplaying **{results\_count}** results of **{results\_count\_total}**[More results...](https://nanotempertech.com)

## Quality in, quality out: How biophysical data powers AI-driven formulation optimization for next-gen biologics

###### Companion post to ourwebinarfeaturing Isabelle Waibel,PhDcandidate, ETH Zurich.

###### by LéaValla— 10 minute read —August 13,2026

###### To learn more about Isabel’s work, watch the fullwebinar

**Watch webinar**

Engineered antibody formats(bispecifics,scfvfragments, nanobodies)are among the most exciting therapeutic modalities in biopharma today.But excitement rarely survives first contact with a developability screen. Unlike the monoclonal antibodies that have been refined over decades, these next-generation formatsdidn'tevolve naturally, and their stability profiles can be unpredictable, variable, and deeply molecule-specific.

The result: formulation scientists are being asked to do more, faster, with less historical data to lean on. Bayesian optimization offers a smarter path forward — but it only delivers when the biophysicaldata feeding it is precise enough to trust.

In a recentNanoTemperwebinar, Isabelle Waibel, aPhDcandidate in the[Arosio Group at ETH Zurich](https://arosiogroup.ethz.ch/), shared how her teamcombined rigorous biophysical characterization with a multi-objective Bayesian framework toidentifyoptimalformulations in just 33 experiments.The central lesson?The quality of the data you put indetermineseverything that comes out.

### Beyond Humira: Thepromise–and theproblem–ofnext-genantibodyformats

For the better part of three decades, monoclonal antibodies (mAbs) like Adalimumab (Humira) have been the backbone ofbiologicsdevelopment. We know how to make them, stabilize them, and formulate them. The field has accumulated a rich body of historical data that makes formulation design, while never trivial, at least navigable.

That foundation is now being tested. Engineered antibody formats includingbispecifics,scFvfragments, and multi-domain constructs offer genuine therapeutic advantages that have the field's attention:

- Improved therapeutic efficiency—Smaller fragments diffuse faster, achieve better tissue penetration, and are more likely to cross the blood-brain barrier.

- Multi-specific targeting—Bispecific and multi-specific formats can engage several molecular targets simultaneously, opening entirely new therapeutic approaches.

- Lower production costs—Many engineered formats can be produced in prokaryote or yeast systems rather than the expensive mammalian cell linesrequiredfor full-lengthmAbs.

But these formats come with a critical liability: theydidn'tevolve. Their structures are engineered, notoptimizedby natural selection, and the mutations that give them their therapeutic properties often come at a cost to intrinsic stability. The field also lacks the systematicdevelopabilitydatasets that exist forcanonicalmabsmaking it far harder to predict where problems willemerge, or how to fix them.

That gap is exactly whatIsabeland her collaborators at Novo Nordisk set out to address.

### 15assays, 73constructs:Here'swhat thedatatoldus

Published in the journalmAbs, [the](https://pubmed.ncbi.nlm.nih.gov/39364796/)[first study](https://pubmed.ncbi.nlm.nih.gov/39364796/)inIsabel's presentation characterized a library of 73 engineered antibody constructs derived from two TNF-alpha-targeting antibodies,Adalimumab andTezunizumab,across 15 biophysical assays. The constructs spanned a wide range of formats, including full-lengthmabs,bispecifics,scFvfragments, andscFv\-scFvtandems.

Here'swhat the data revealed:

###### Thermal stability

###### Melting temperature (Tm) and onset of aggregation upon heating were measured using[Prometheus Panta](https://nanotempertech.com/prometheus-panta/).Most formatsexceededthe 60°C threshold that characterizes most marketed antibodies. But variation within format families was significant, showing that the specific sequence matters as much as the format category.

###### Colloidal stability

###### The diffusion interaction parameter (kD)wasalso measured by Prometheus Pantaandtold a more differentiated story. Small fragments, particularlyscFvandscFv\-scFvconstructs, showed the most pronounced colloidal stability challenges. Critically, engineered formats showed far greater within-groupvariation thancanonicalmAbs, underscoring how sensitively these formats respond to individual mutations.

###### Long-term stability (12 months)

###### At 30°C, most variants remained stable, thoughscFv\-scFvfragments showed some aggregation. At 40°C, full-lengthmabsoutperformed fragments andbispecificsacross the board, and significant fragmentation wasobservedandattributedlargely toGS linkers, which can beoptimizedfurther.

###### The flag-based risk ranking

###### To make the complexity of 15 assays actionable, the teamadopted a "flag" system: each time a molecule exceeded a risk threshold for any measured property, it received a flag. The more flags, the higher the overall developability risk.

###### The results were instructive. Canonicalmabsaccumulated the fewest flags,confirming their relative robustness. But the data also showed that format family is not destiny: somebispecificsandscFvvariants performednearly as well astraditionalmabs, whilescFv\-scFvtandems showed the highest risk across the board.

The key finding:Engineered antibody formats carry higher developability risk than traditional mabs, but the specific liabilities vary widely by format and by molecule. That variability is precisely why tailored, data-driven formulation strategies are essential.

Asa final step, the team tested whether thein silico[tool TAP (Therapeutic Antibody Profiler)](https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/tap)could predict these developability profiles from sequence alone. It performed well forcanonicalformats but struggled with the novel constructs,reinforcing thatthe field needs more experimental data to train andvalidatein silico tools for engineered modalities.

### Fromscreeninghundreds offormulations tofinding theoptimumin 33experiments

With the developability liabilities of their model protein,bocutizumab,clearly mapped, the next question was: how do you fix them, without running hundreds of experiments youcan'tafford?

The answerIsabel’s team developed: [a multi-objective Bayesian optimiz](https://nanotempertech.com/nanodsf/https://pubmed.ncbi.nlm.nih.gov/41002022/)[ation framework](https://nanotempertech.com/nanodsf/https://pubmed.ncbi.nlm.nih.gov/41002022/).

###### What are the currentchallengeswithtraditionalDesignof Experiments(DoE)?

###### ConventionalDoEapproachessuch ascentral composite designorBox-Behnken design,fix the experimental plan at the outset. You define your variables and levels upfront, run the full matrix, and analyze the results. This worksreasonably wellfor simple, well-understood systems. But for complex biologics in large design spaces, it has two significant weaknesses:

- Local optima:With a pre-fixed plan,DoEhas no mechanism to redirect experiments toward areas that look promising. You can easily converge on a locally good formulation while missing the globaloptimumentirely.

- Rigidity:The method cannot learn from early results and adapt. Every experiment is equally uninformed.

###### HowdoesBayesianoptimizationwork?

###### Bayesian optimization is a sequential, adaptive approach tosearchingcomplex design spaces.Here'show it works in practice:

- Initialize:Run a small set of experiments (13 in this study) covering the designspace broadly.

- Train:Use the results to train surrogate models:in this case, Gaussian processes,that learn the relationship between formulation variables and stability outcomes.

- Suggest:The model recommends the next experiment by balancing two strategies: Exploitation (75%): Focus on regions of the design space that have already shown promise, guided by theParetofront of the surrogate models. Exploration (25%): Send experiments into novel, uncharted regions to ensure global coverage and avoid local optima.

- Repeat:Measure the new formulations, retrain the models, and iterate.

To make this experimentally practical, the team used the Kriging Believerapproach,retrainingthe model four times with simulated results to generate five new formulation suggestions per round, all of which can be prepared and measured in parallel.

The optimization simultaneously improved three properties: thermal stability (Tm), colloidal stability (kD), and resistance to agitation-induced aggregation (remaining monomer after shaking). Osmolality was incorporated as a hard constraint, keeping all generated formulations within the physiologically tolerable range of 100–600mosm/kg.

###### The Results

###### After 33 total experiments(13 for initialization, then four rounds of five)the algorithmidentifiednear-Paretooptimalformulations.That'sless than half the experimentcountof comparabledoemethods, and crucially, it comes with a guarantee of global search coverage that staticdoemethods cannot offer.

###### In terms of speed: TmandkDmeasurements on Prometheus Panta take under one hour per round for up to 48 samples in parallel. The rate-limiting step in this study was the agitation assay, which requires one week of shaking at 1 mg/mL.With alternative interfacial stress assays, even that bottleneck could be removed,making the full optimization achievable in a single day.

### "Garbagein,garbageout":Whymeasurementprecisionmakes orbreaks themodel

Bayesian optimization isonly as powerful as the data that feeds it. The surrogate models learn from experimental results,which means measurement noise must besubstantially smallerthan the effect of the excipients being tested. If your instrumentcan'treliably distinguish a 0.5°C shift in Tm, neither can your model.

This is where instrument precision becomes a strategic asset, not just a technical specification.

In this study, Prometheus Panta delivered:

- Tmprecision: An average 95% confidence interval below 0.1°C,one of the tightest reported for[nano](https://nanotempertech.com/nanodsf/)[DSF](https://nanotempertech.com/nanodsf/)TM\-based thermal unfolding measurements.

- kDprecision: Confidence intervals were wider, as expected for interaction parameter measurements, but remained within acceptable range. The most reliable results wereobservedat clearly high or clearly lowkdvalues, where the[DLS](https://nanotempertech.com/dls/)curve slope is most distinct.

Beyond raw precision, a practical advantageshapedthe entire workflow: Tm andkDcan be measured from the same sample, in the same Prometheus Panta run. For a study where material conservation and throughput bothmatter, this matters. With capacity for up to 48 samples in parallel, teams can run full rounds of optimization(including replicates across multiple conditions)without scheduling separate instrument blocks.

##### What was especially useful for meisthat I could measure the melting temperatureand alsotheKdwith the same sample in the same instrument.Sothis saved me a lot of timeand alsomaterial.

#### Isabel Waibel

Scientist, ETH Zürich

This is the practical meaning of "quality in, quality out." The machine learning modeldoesn'tknow your instrument. It only knows the numbers you giveit. Prometheus Panta's measurement precision is what makes those numbers trustworthy,and that trustworthiness is what makes the entire optimization framework function.

### Thesurprisingstory thedatatoldabout Arginine

Beyondidentifyingoptimalformulations, the Bayesian approach generated a rich dataset that shed light on why the algorithm made the choices it did,and some of the findings challenged conventional formulation wisdom.

pHconverged at ~6. This reflects a fundamental trade-off: higherpHtends to improve thermal stability (Tm), while lowerphdrives better colloidal stability (kD) by moving the molecule away from its isoelectric point and increasing intermolecular repulsion. The algorithm balanced both by consistently converging nearpH6,closelyaligned with thepHrange (~5.8) seen in most commercially approvedmAbformulations.

Sorbitol settled at ~300mm. Intermediate-to-high sorbitol concentrations consistently improved Tm,a well-characterized effect since the 1980s, explained by the preferential exclusion of polyols from the protein surface, which promotes a more compact and thermodynamically stable protein structure.

Arginine was consistently excluded. This is the headline finding. Arginine is one of the most widely used excipients in biopharmaceutical formulation,and yet, for this molecule, the algorithm avoided it innearly everyoptimization round. Across all three target properties, arginine had a negative effect.

Thelikely mechanisms: at formulation-relevantpH, arginine is positively charged, significantly increasing ionic strength and reducing the repulsive intermolecular interactions thatmaintaincolloidal stability. Its guanidinium group also has known chaotropic properties that can destabilize protein structure, contributing to reduced Tm.

This isprecisely the kind of system-specific insight thatprior-knowledge\-based formulation design misses. Arginine works well for many molecules. ForBocutizumab, itdoesn't,and only a data-driven approach that interrogates the actual behavior of the actual molecule would have revealed that.

The broader lesson:Optimal formulation is not generic.It'smolecule-specific, and it requires the kind of high-quality, systematic experimental data that biophysical tools like Prometheus Panta are built to generate.

### Whatthismeans foryourformulationprogram

The ETH Zurich and Novo Nordisk collaborationdemonstratessomething important: the convergence of biophysical measurement and machine learningisn'ta future ambition.It'sa workingmethodologytoday.

Here are the key takeaways:

- Engineered antibody formats need formulation strategies as novel as the formats themselves. Historicalmabformulation knowledge is a starting point, not a solution.

- Bayesianoptimization finds global formulation optima with a fraction of the experimental effort of traditionalDoE,without the risk of converging on a localoptimum.

- High-precision biophysical data istheessential input. Instruments like Prometheus Panta that deliver low measurement noise and multi-parameter output from a single sample are direct enablers of better ML models.

- Multi-property optimization is achievable. Thermal stability, colloidal stability, and interfacial stability can all be improved simultaneously,and trade-offs can be navigated algorithmically rather than by intuition.

- The dataset you generate has value beyond the optimization itself. It becomes a mechanistic resource for understanding how individual excipients behave with your specificmoleculewhich isan asset for future programs.

Want to learn more about Isabel’s work: read the publications.

[https://pubmed.ncbi.nlm.nih.gov/39364796/](https://pubmed.ncbi.nlm.nih.gov/39364796/)

[https://pubmed.ncbi.nlm.nih.gov/41002022/](https://pubmed.ncbi.nlm.nih.gov/41002022/)[https://pubmed.ncbi.nlm.nih.gov/41002022/](https://pubmed.ncbi.nlm.nih.gov/41002022/)

### Frequently Asked Questions (FAQ)

##### 1.How many experiments does Bayesian optimization require for protein formulation development?

###### In this study, 33 experiments(13 for initialization and 4 rounds of 5)were sufficient toidentifynear-Paretooptimalformulations. This is less than half the number required by comparabledoemethods such as central composite or Box-Behnken designs, and it comes with global search coverage that static approaches cannot guarantee.

##### 2.What biophysical assays are used in AI-driven formulation optimization?

###### The three target properties in this workflow were thermal stability (melting temperature, Tm, measured bynanoDSF), colloidal stability (diffusion interaction parameterkD, measured by DLS), and agitation resistance (remaining monomer after one week of shaking, measured by SEC or similar).All three were chosen for their relevance to real-world developability risk.

##### 3.Can Bayesian optimization be applied to any protein therapeutic?

###### Yes. The design space(excipient types, concentration ranges,pHwindows)is fully customizable for each molecule. The approach is especially valuable for novel modalities like engineered antibody formats, where historical formulation data islimitedand molecule-specific behavior is hard to predict from prior knowledge alone.

##### 4.How long does one round of Bayesian optimization take with Prometheus Panta?

###### TmandkDmeasurements together take less thanone hour per round for up to 48 samples in parallel on Prometheus Panta. In this study, the rate-limiting step was the agitation assay, whichrequiredone week of shaking. With alternative assays,such as nanoparticle-based interfacial stress assays,that bottleneck can be removed, making the full optimization achievable within a single day.

##### 5.Is AI just a buzzword in formulation development, or does it deliver real value?

###### It delivers real value,but only whenit'sgrounded in high-quality experimental data.AsIsabel noted in thewebinar, machine learning isultimately statistics: the outputs are only as reliable as the inputs. The combination of rigorous biophysical measurement and a well-designed adaptive algorithm is what makes the difference between a genuinely optimized formulation and an expensive guess.

### Ready tobringqualityin,andgetqualityout?

The era of running hundreds of formulation experiments to find a good-enough answer is ending.What'sreplacing it is a smarter approach: adaptive, instrument-enabled, and designed to find the right answer,not just a workable one.

Prometheus Panta is builtfor exactlythis kind of workflow. Its combination ofnanoDSF\-based thermal stabilitymeasurement and DLS-based colloidal stability characterization, from the same sample in the same run, with the throughput and precision that machine learning demands, makes it the biophysical foundation for next-generation formulation development.

**Watch the full webinar recording**

Learn more about how Isabelle Waibel's team built andvalidatedthis approach, including the full design space setup, model training details, and excipient-effect analysis.

**Explore Prometheus Panta**

See how Prometheus Panta supports formulation development from early candidate screening through late-stage process characterization.

**More blogs on AI Drug Discovery**

Read more about how AI and ML models support drug discovery workflows, and how NanoTemper's instrument can easily integrate to train and validate them.

##### There's more to explore

###### Discover more science-led content, practical guidance, and resources to deepen your knowledge and move your research forward with confidence.

**Back to resource centre**

Breakthrough faster. Together. [Let's talk](https://nanotempertech.com/contact-specialist/) newsletter Get our latest publications, webinars, and case studies. [https://nanotempertech.com/](https://nanotempertech.com/) We develop biophysical tools and reagents for discovery, development, and beyond. Helping teams to breakthrough faster, together. Explore Explore [Resources](https://nanotempertech.com/resources/) [About](https://nanotempertech.com/about/) [Careers](https://nanotempertech.com/careers/) [Contact](https://nanotempertech.com/contact-specialist/) Solutions [Biologics](https://nanotempertech.com/biologics/) [TPD](https://nanotempertech.com/tpd/) [Membrane Protein](https://nanotempertech.com/membrane-protein/) [Products](https://nanotempertech.com/#solutions) [Consumables](https://shop.nanotempertech.com/) [Support](https://nanotempertech.com/support/) Technologies [nanoDSF](https://nanotempertech.com/nanodsf/) [DLS](https://nanotempertech.com/dls-7/) [SLS](https://nanotempertech.com/sls-6/) [Backreflection](https://nanotempertech.com/backreflection/) [Spectral Shift](https://nanotempertech.com/spectral-shift/) [MST](https://nanotempertech.com/mst/) [TRIC](https://nanotempertech.com/tric/) [LinkedIn](https://www.linkedin.com/company/nanotemper-technologies-gmbh) [YouTube](https://www.youtube.com/user/NanoTemper) © 2026 NanoTemper Technologies. All Rights Reserved. [Imprint](https://nanotempertech.com/imprint/) [Privacy Policy](https://nanotempertech.com/privacy-policy/)