The Translational Divide and Digital Biomarker Validation Part 2: Cross-Disciplinary Collaboration & Future Applications in AI and Digital Measures

What happens when preclinical and clinical researchers actually collaborate? A smarter, more translatable approach to digital biomarker validation. In Part 1, we explored the problem—now, let’s talk about the solution. From AI-driven pathology to digital behavioral biomarkers, cross-disciplinary collaboration is the key to unlocking forward and reverse translation. If we align validation efforts, we can accelerate drug development and improve regulatory acceptance of digital measures. Here’s how we get there.

Cross-Disciplinary Collaboration in Action: Bridging the Translational Gap 

The development of the in vivo V3 validation framework is a prime example of impactful cross-disciplinary collaboration. It shows that when experts from animal research and human research actually talk to each other (preferably over good tea), great things can happen. Let’s break down why this collaboration worked and what it means going forward: 

A Common Framework = Common Understanding: By using Digital Medicine Society (DiMe) clinical validation framework as a starting point, the The 3Rs Collaborative (3RsC) Translational Digital Biomarker Initiative team with colleagues from Digital In Vivo Alliance created an instant bridge between disciplines (Validation framework for in vivo digital measures - PubMed ). Clinical researchers immediately recognize the V3 concept, and preclinical researchers get a crash course in what kind of evidence the clinical folks expect. This alignment is golden because one big barrier in translational science is simply misalignment – people measuring different things or validating in ways that don’t correspond. Now, with a shared framework, a validated digital biomarker in an animal has a clearer path to being accepted as relevant in a clinical context. 

Holistic Stakeholder Involvement: The initiative didn’t happen in a vacuum. It involved pharma companies, biotech startups, academia, and even technology vendors (Validation framework for in vivo digital measures - PubMed). Each brought something to the table. The industry scientists know what will actually get a drug program excited, the technologists know what the devices can or can’t do, the academics bring deep expertise on the animal models, and regulators (even if informally involved) hint at what evidence they’d want to see. It’s like assembling the Avengers of digital biomarker validation – every perspective helps cover a weakness. The result is a framework that’s scientifically rigorous and also practical for real-world R&D use. 

Focus on Translational Relevance: The collaboration kept eyes on the prize – improving translation. The emphasis on making animal data more translatable to human outcomes is woven throughout the framework. By sharing this focus, the preclinical folks and clinical folks were essentially working on the same problem, just from different angles. It’s a subtle but important shift from “my research vs. your research” to “our translational problem”. When everyone’s invested in solving the same problem (like why a certain neurobehavioral biomarker isn’t carrying over from mice to humans), collaboration becomes natural. 

The success of the in vivo digital measures validation framework gives a blueprint for tackling other translational gaps. It shows that rather than throwing fancy tech or more data at the translational disconnect, sometimes the answer is getting the right people in a (virtual) room and hashing out a framework that forces integration of knowledge. The next sections will look at where else we can apply this recipe of bidirectional learning and cross-talk, especially with the rise of AI and novel digital measures. 

Beyond Biomarkers: Opportunities for Bidirectional Learning with AI and Digital Measures 

The principle of two-way translation and collaborative validation isn’t limited to activity monitors in mice or wearables in humans. There are plenty of other areas where aligning preclinical and clinical approaches – and involving multidisciplinary teams – could pay off big. Here are a few intriguing opportunities where AI and digital measures could benefit from similar forward-reverse learning and unified validation frameworks: 

• Digital Pathology and AI-Assisted Histology: Pathologists in drug development often examine tissue slides from animal studies (to check for toxic effects, for example) and from human trials (biopsies, etc.). AI has entered the scene with algorithms that can grade tumors or identify cell types on images. A bidirectional approach here might involve validating an AI model on both animal and human tissues in tandem (keeping in mind that there are similarities e.g. Hepatocyte Structure and differences e.g. Lobule Size & Structure). For instance, an algorithm that quantifies liver fibrosis could be trained on annotated rodent liver sections and human biopsy images. Forward translation: ensuring the AI’s features (say, collagen density measures) correlate with disease stage in animals and then seeing if the same holds in patients. Reverse translation: taking patterns noted in patient tissues (perhaps a novel imaging biomarker) and checking if the animal model exhibits those, to confirm the model’s relevance. A joint validation framework could specify that any AI pathology biomarker needs to be verified on technical accuracy (image scanning resolution, etc.), analytically validated (algorithm consistency on known histological features) and biologically validated in both species. This might streamline how we qualify AI tools for regulatory submission in both preclinical and clinical contexts simultaneously. 
• Behavioral and Neurological Biomarkers Across Species: Neurological disorders are notorious for translational hurdles – an experimental drug might make a mouse maze champion, but a human with Alzheimer’s still struggles with memory. With digital biomarker tools, we can measure behavior and physiology more continuously and objectively in both animals and humans. For example, consider sleep and circadian rhythm data. In humans, wearables and home IoT devices can track sleep duration, sleep fragmentation, and activity cycles. In mice, home-cage sensors can do the equivalent ( Emerging Role of Translational Digital Biomarkers Within Home Cage Monitoring Technologies in Preclinical Drug Discovery and Development - PubMed). A collaborative framework could validate that a “sleep fragmentation index” derived from a mouse correlates with known physiological stress markers in that mouse (biological validation) and likewise that a similar index from human wearables correlates with patient outcomes (clinical validation). If both hold true, you’ve got a powerful biomarker that’s comparable across species. AI can assist by finding subtle patterns (like micro-movements during sleep that predict seizure risk, for instance) and those patterns can be checked in animal models (forward) and then fed back if found in patient data (reverse). The key is designing studies in parallel – perhaps a clinical study and an animal study both collecting the same type of sensor data – and then jointly analyzing them. 
• AI in Drug Safety Monitoring: Digital measures are not just for efficacy; they’re crucial for safety. Imagine using smart cages to detect subtle signs of adverse effects in animals – decreased spontaneous activity might hint at malaise, or changes in gait might hint at emerging neuropathy in a rat. In the clinic, patients could wear sensors or use apps to report similar subtle symptoms. A bidirectional validation approach could unite these efforts. One concrete example: detecting drug-induced balance issues. An AI algorithm might analyze video of a mouse on a beam or ladder test to quantify balance. Meanwhile, in a trial, an app might use the phone’s accelerometer to measure patient balance during daily activities. By sharing notes, researchers could ensure that the algorithm is verified (it accurately measures wobble in both mice and humans), analytically validated (the balance metric correlates with known balance tests or markers in each species), and clinically/biologically validated (in animals, the metric predicts neuropathy pathology; in humans, it predicts fall events or clinical neuropathy outcomes). This way, a safety signal detected in animals is directly translatable to what to watch in patients, and vice versa. 
• Cognitive and Psychiatric Digital Endpoints: Assessing cognitive function or mood is hard enough in humans, let alone animals. But digital tech is offering new windows into these domains – from computer-based cognitive tests, smartphone sensing of social behavior, to machine learning analysis of rodent vocalizations or nest-building as proxies for well-being. Cross-disciplinary teams could develop composite digital biomarkers for, say, anxiety that work in both mice and humans. If a mouse’s movement patterns in an open field test and a human’s patterns on a smartphone anxiety app are both being measured, an integrated validation might reveal common signals (like fragmented bouts of activity or specific circadian disruptions). Perhaps AI finds that both anxious mice and anxious humans show disrupted day-night activity cycles. Forward translation would encourage the human researchers to look for that pattern, reverse translation would encourage the animal researchers to confirm their model shows the pattern seen in clinic. By validating the digital measure (activity cycle regularity) in both contexts, we increase confidence that treating the mice to normalize the pattern is relevant, and observing it in humans is meaningful. 

In all these examples, the recipe is familiar: get everyone together early (before everyone runs off doing separate validations), use a unified framework so data can be compared or combined, and let the questions and results flow both ways between the animal lab and the clinic. It’s easier said than done – different studies have different constraints – but the 3RsC’s success with the in vivo digital measures framework gives a roadmap. They showed that starting with a high-level framework (like V3 framework) and then refining it for the domain can align efforts without stifling innovation. 

My final thoughts  

Achieving translation in biomedical research is a bit like teaching two different species to dance together. It takes patience, understanding, and a good choreography – but when it works, it’s a sight to behold. The collaboration between the 3Rs Collaborative TDB initiative and DiMe and authors from DIVA has yielded a scientifically rigorous validation framework that brings preclinical digital biomarker validation up to clinical standards. It’s a framework that not only holds devices and algorithms accountable for quality and relevance, but also implicitly nudges researchers to think in translational terms – asking “does this measure in my animal matter for human health?” at every step. 

I highlighted how this framework builds on DiMe’s V3 pillars of Verification, Analytical Validation, and Clinical (now biological) Validation, and how it exemplifies cross-pollination between disciplines ( Validation framework for in vivo digital measures - PubMed ). Beyond this specific case, we discussed why forward and reverse translation are two sides of the same coin in digital biomarker work – and indeed in much of biomedical innovation ( Application of Machine Learning in Translational Medicine: Current Status and Future Opportunities - PMC ) Genentech: Reverse Translation). By learning from each other, preclinical and clinical researchers can avoid reinventing wheels and, importantly, avoid the dreaded scenario of a “great” animal result that means zilch in humans or a critical human observation that never makes it back to the lab – a missed reverse translation opportunity (Digitalization of toxicology: improving preclinical to clinical translation - PubMed). 

The future is bright for those willing to embrace this bidirectional dance. I envision a world where every digital measure, every AI model, and every novel biomarker is developed with an eye on both the microscope and the patient monitor. Frameworks like the one from 3RsC with colleagues from DIVA provide the sheet music – a common melody that different players in the orchestra can follow. 

So whether you’re a bench scientist training mice to run on tiny treadmills, or a clinician scientist figuring out how to validate a patient’s smartphone gait tracker, remember that you’re ultimately working on the same puzzle. And sometimes, the piece you need might just be on the other side of the translational fence. It might take a bit of humility (and perhaps an extra Zoom meeting or two) to ask your counterparts what they’re seeing, but the payoff – more robust science, faster development of therapies, and digital tools that we can actually trust – is well worth it. 

In the end, bridging the gap isn’t just about frameworks and meetings; it’s about mindset. A mindset that every preclinical experiment is done with a vision of the patient in mind, and every clinical study is designed with an appreciation of the biology that underpins it. The 3Rs Collaborative with colleagues from DIVA and DiMe have set a strong example. It’s up to the rest of us to follow suit, perhaps with our own twist of humor and humanity along the way. As we move forward, let’s keep the conversations flowing across disciplines – after all, the only way to cross the “valley of death” in translation is to build a bridge sturdy enough for everyone to walk across together.