Quotient Sciences Advances AI-Guided Drug Formulation to Shorten Clinical Development

16 November 2026 | Monday | Interaction

An integrated approach combining formulation development, manufacturing and clinical testing aims to reduce development timelines, improve drug performance and support earlier clinical decisions.
Editor: BioPharma Boardroom

Editor: BioPharma Boardroom

Developing effective drug formulations remains a major challenge for biopharma companies, particularly for molecules with poor solubility or complex release requirements. Quotient Sciences is exploring how machine learning and integrated manufacturing and clinical testing can improve formulation decisions while reducing development time and costs. Interim clinical results showed that its algorithm reached a predefined pharmacokinetic target within three dosing periods. BioPharma Boardroom speaks with Dr. Andrew Lewis, Chief Scientific Officer, Quotient Sciences about AI-guided formulation development, integrated CRDMO services, manufacturing challenges and its collaboration with Acesion Pharma.

Q. How is AI-enhanced formulation development changing the speed and accuracy of early-stage clinical decision-making, and what advantages does it offer over traditional approaches?

The hardest prediction in early formulation work has always been the jump from the dissolution bath to the patient. Teams can generate good in vitro data quickly. Knowing how a composition will translate into human pharmacokinetics has traditionally meant several rounds of making, dosing to volunteers, evaluating the results and reformulating, and that can take months or years.

What has changed is that we can now test whether a model can learn that relationship directly. In our recent clinical study, we used a proprietary ML algorithm to direct the formulation optimisation towards pre-defined pharmacokinetic targets. Earlier laboratory work had already shown it could map the formulation design space after screening a third fewer formulations than conventional methods, and we hypothesised that if it could learn the relationship between composition and in vitro performance, surely it could learn the relationship between composition and in vivo performance. It entered the trial trained only on in vitro release data. After each dosing period it was retrained on dissolution and pharmacokinetic results, then selected the next composition and dose. Compositions proposed by the algorithm were reviewed by a Safety Advisory Committee before testing to ensure they would be safe for the volunteers. Interim results show it reached the preset pharmacokinetic target within three dosing periods. 

For sponsors, the bigger advantage is not raw speed but where the evidence lands. Each clinical period becomes a learning step rather than a checkpoint, and an in silico model on the relationship between composition, in vitro and in vivo performance built as the data is generated and continuously refined. Decisions on which drug product to progress are made with human data in hand, earlier, and with less API spent getting there.

Q. What are the main challenges in integrating drug development and manufacturing, and how can an integrated CRDMO model help reduce hand-offs and shorten development timelines?

Most early programmes don't lose time in the lab or the clinic. They lose it in the gaps between them. A formulation is developed by one group, transferred to a manufacturer, released, shipped, held on stability and only then dosed by a clinical provider. Each hand-off adds contracts, technology transfer and waiting, and each one dilutes ownership of the science.

An integrated model goes after that white space. When drug product manufacture and clinical dosing sit within one organisation, the interval between making a batch and dosing it can fall from months to days. Product is made to clinical need, so batch sizes and stability requirements shrink. An independent Tufts CSDD analysis of Translational Pharmaceutics® put the average saving at more than 12 months of development time.

What I'd stress to sponsors is that time is only part of the return. Removing hand-offs means the people who designed the formulation see the human data and can act on it within the same study. That feedback loop is where better decisions come from. It is also, increasingly, what makes AI useful, because a model learns far more from connected data than from fragments passed between vendors.

Q. For challenging molecules with poor solubility or modified-release requirements, what formulation strategies can help improve drug performance while keeping development programmes on track?

The mistake I still see most often is committing to one enabling technology before the molecule has told you what the problem really is. Poor exposure can be driven by solubility, dissolution rate or permeability, and each calls for a different answer. Using a framework such as the Developability Classification System to diagnose the driver first saves a great deal of wasted formulation effort.

The second lesson is to keep options open into the clinic rather than choosing in the lab. In one programme, a poorly soluble compound showed low exposure and a large food effect in its first human study. Three approaches, a micronised API, a self-emulsifying lipid system and a spray-dried dispersion, were screened side by side in healthy volunteers. A new lead formulation was identified in about six months.

Modified release benefits from the same thinking. Building a design space into the regulatory submission lets you adjust release rate and dose between dosing periods based on the pharmacokinetics you are actually seeing. That is precisely the territory our recent AI study has been working in, and it suggests the number of clinical rounds needed to hit a target profile will keep falling.

Q. How can biopharma companies build more flexible development and manufacturing strategies to respond to changing clinical results, regulatory requirements and commercial priorities?

Flexibility is usually treated as a contingency, something you hope you won't need. In early development I'd argue it should be designed in, because at this stage of development relatively little is known about the molecule and the drug product and surprising results are relatively common. 

There are three practical levers. The first is regulatory: protocols and submissions written with flexibility built in through adaptive designs enable decisions to be made on the emerging clinical data. The second is supply. Manufacturing in real time to clinical need, rather than building large batches months ahead, reduces stability requirements and avoids being locked into a formulation the data no longer support. The third is study design. Adding patient cohorts to an integrated programme can bring proof of concept closer within a single clinical study.

What's newer is the role of data. A model that links composition to in vitro and human performance becomes a reusable asset. If the target product profile shifts, perhaps because a once daily regimen becomes commercially essential, you are not starting from zero. You are asking a better question of evidence you already hold.

Q. How are outsourcing priorities evolving, and what should biopharma companies consider when selecting a CDMO or CRDMO partner for long-term development and manufacturing needs?

The conversation has shifted. A decade ago, many sponsors outsourced task by task, choosing the cheapest or most available provider for each step. Today the questions I hear are about continuity: who will still own this programme when it moves from first in human to patient studies, and how much risk sits in each transfer between vendors.

Funding pressure has sharpened that. When capital is tight, a lost quarter matters, and so does the judgement of the people running the work. In my experience, the scientific acumen of a provider and the quality of what it produces can decide whether a programme succeeds.

My advice is to look beyond the service list. Ask how a partner moves from formulation to clinic, and how many contracts, protocols and regulatory submissions that involves. Sponsors increasingly value a single contract and unified protocol from first in human through to proof of concept for exactly this reason. Then ask about data. Where it lives, who owns it and whether it can sharpen future decisions on the same molecule is now a strategic question, particularly as AI tools depend on it.

Q. Following Quotient Sciences' recent work on AI-enhanced formulation development, including its collaboration with Acesion Pharma, what lessons can the industry take from these initiatives, and how do you see AI influencing drug development in the coming years?

The most useful thing our recent work has shown is where AI earns its place: not in replacing experiments, but in choosing them. In our clinical study, interim data show the algorithm that started trained on in vitro data alone, learned from each dosing period, and reached its pharmacokinetic target in three attempts. Each human data point improved the next decision.

That only works when the loop is short. A model that waits months for data passed between vendors learns slowly. One sitting inside integrated manufacturing and clinical testing learns at the pace of the study, and is the perfect synergy to get the most out of the technology.

This is however just the start, and more work needs to be done to prove the technology.  Our study used a well characterised generic drug, and we are now applying the approach to novel molecules with Acesion and other customers. Small datasets limit how far a model can extrapolate, and expert judgement on design space and safety still frames every decision.

Over the next three to five years, I expect the digital twin built in early development to travel with the molecule, informing scale up, line extensions and regulatory discussion. The prize is not speed alone but fewer wrong turns.

 

Survey Box

Poll of the Week

Which area of biopharmaceutical research excites you the most?

× Please select an option to participate in the poll.
Processing...
× You have successfully cast your vote.
 {{ optionDetail.option }}  {{ optionDetail.percentage }}%
 {{ optionDetail.percentage }}% Complete
More polls
Stay Connected

Sign up to our free newsletter and get the latest news sent direct to your inbox

© 2026 Biopharma Boardroom. All Rights Reserved.

Show

Forgot your password?

Show

Show

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close