Researchers are increasingly turning to modular mechanistic models to unlock greater efficiency and robustness in mRNA manufacturing, offering a more flexible way to optimize in vitro transcription (IVT) while reducing costly experimental work. According to Wei Xie, PhD, associate professor of mechanical and industrial engineering at Northeastern University, and her colleagues, modular approaches can increase productivity and product quality.
“A modular modeling approach simplifies the complex IVT reaction network by dividing it into discrete, reusable, mechanistically defined steps,” Xie said. “This structure improves mechanistic understanding by clarifying how each step impacts key quality attributes, including yield, capping efficiency, and transcript integrity.”
Rather than relying on a single monolithic model, the framework separates IVT into individual components, such as initiation, elongation, and termination, as well as parallel processes including mRNA degradation and precipitation. Each module can be independently calibrated, validated, and refined as new experimental data become available, allowing researchers to continuously improve predictive performance without rebuilding the entire model.
The modular architecture also lends itself to the evolving nature of mRNA therapeutics. Because the framework mirrors the modular structure of nucleic-acid sequences, it can be rapidly adapted for new constructs, accelerating process development for emerging vaccines and therapeutic candidates while minimizing redevelopment effort. Beyond improving process understanding, the model provides a powerful diagnostic platform for identifying production bottlenecks that constrain yield, productivity, or product quality.
The framework combines Shapley value-based sensitivity analysis, residual analysis, and simulated reaction trajectories to pinpoint limiting process variables. Sensitivity analysis identifies parameters with the greatest influence on performance, while comparisons between predicted and experimental results reveal missing mechanisms or model deficiencies. Simulated reaction profiles can also highlight issues such as nucleotide depletion or suboptimal magnesium-to-nucleotide ratios before they become significant manufacturing challenges.
“Together, these tools provide a data-driven, mechanistic approach to quickly diagnose constraints and guide targeted process optimization,” Xie explains.
The approach also offers significant advantages during scale-up, one of the most challenging phases of bioprocess development. Because the model is grounded in fundamental molecular reactions and biochemical mechanisms rather than empirical correlations, it maintains predictive capability across different manufacturing scales and can be readily applied to new mRNA sequences, all without extensive redevelopment.
Xie says the framework supports predictive design of scale-dependent control strategies, including dynamic pH regulation and fed-batch nucleotide feeding schemes, helping manufacturers reduce development timelines while improving process robustness during technology transfer.
“A key advantage of the modular architecture is its flexibility and interoperability” Xie says. “New enzymes, reagents, or process steps can be incorporated by simply updating or adding the relevant module, without recalibrating the entire model. The framework’s ability to accommodate heterogeneous datasets generated under varying process conditions further supports rapid evaluation of manufacturing innovations while maintaining model consistency.
Perhaps the greatest impact of Xie’s approach lies in advancing quality-by-design (QbD). Acting as an in silico development platform, the modular model enables researchers to evaluate process variables before entering the laboratory. Coupled with digital twin-based Bayesian optimization, the platform narrows the experimental search space, reducing trial-and-error studies while conserving expensive reagents, such as T7 RNA polymerase.
As mRNA pipelines continue to expand beyond vaccines into broader therapeutic applications, modular mechanistic modeling is emerging as a valuable digital bioprocessing tool, enabling manufacturers to accelerate development, strengthen process understanding, and deliver more consistent product quality with fewer experimental resources.
The static, mounted monitoring systems currently used inside fermentation vats are poised to be replaced in the near future with a network of free-floating bioelectronic sensors, if the vision of researchers from Boston University and Capra Biosciences reaches fruition.
Designed for both vat and continuous bioprocessing systems, this bioelectronic sensor network could, ideally “provide spatial information about where they are in a heterogeneous bioreactor platform…as well as multiple measurements of things such as temperature, pH, dissolved oxygen, and dissolved carbon dioxide,” Rabia Yazicigil, PhD, associate professor, Boston University (BU), and lead principal investigator for this project, tells GEN.
Consequently, the network will report data that enable biomanufacturers to determine whether the solution is mixing properly and to identify transit times throughout the process, in addition to specific processing parameters.
“The key innovation…is that these systems integrate living cells into the electronics,” Miguel Jimenez, PhD, assistant professor, BU, emphasizes. The inclusion of microbes—bacteria or yeast cells, for example—“supercharges the sensors,” enabling them to monitor more parameters that are directly relevant to biomanufacturing.
Roughly the size of a chickpea, these sensors never leave the bioreactor. “That allows us to get measurements throughout the reactor… which helps us build a really rich data set that we can then feed into models to help us monitor and predict performance,” notes Mark Poole, PhD, senior director of manufacturing and applied AI, Capra Biosciences.
Paradigm-shifting potential
“Having lots of high-quality measurements at different points in the reactor is game-changing for any biomanufacturing company,” Poole says.
Jon Valdez, program manager at BioMADE, which funded the project as part of a $21.4 million investment in 14 projects to advance the bioindustrial manufacturing industry, agrees, calling it potentially paradigm-shifting. Potential applications extend to clinical monitoring—where a prior collaboration focused on human gut monitoring. The technology is solvent-agnostic but may be most effective in a water-based environment, enabling applications that may include soil and water quality monitoring. Benefits, he says, include lower costs per sensor (estimated at $10−$100) and decreased risk of contamination.
The project is two-tiered. The first tier, the electronics-only sensor, is the nearest to commercialization. Industrial-scale testing will be conducted soon at Capra facilities. “That [alone] would signify a big advance,” Jimenez says, citing the ability to field networked sensors capable of measuring multiple conditions throughout a bioreactor or continuous production process.
The second tier adds the bio component to those sensors. This feature is in academic development. Primary challenges are how to design biohybrid sensors that can be autoclaved or cleaned-in-place, and strategies to stabilize and encapsulate the microbes to be compatible with industrial requirements. The researchers are considering possible approaches now.
AI could help cell and gene therapy manufacturers gain a deeper understanding of the complex production processes used to make their products and predict problems before they occur.
A team led by researchers at Northeastern University College of Science in Boston made the case for AI use in a recent paper, arguing that the variability inherent in cell and gene therapy production can be difficult to manage using conventional tech.
Lead author, Jared Auclair, PhD, dean of the College of Professional Studies at Northeastern, tells GEN, “Unlike monoclonal antibodies or recombinant proteins, cell and gene therapies are living or highly complex biological products, making them inherently more variable and difficult to manufacture consistently.
“Every step, from sourcing starting material to manufacturing, analytical testing, storage, and delivery, can influence the final product,” he adds.
Understanding complex, multi-parameter interactions is exactly the sort of challenge at which AI excels, Auclair says, citing the ability to identify critical process attributes as an example.
“AI has the potential to transform cell and gene therapy manufacturing by moving from reactive to predictive manufacturing. Machine learning can optimize process parameters, predict batch failures before they occur, enable digital twins to simulate manufacturing changes, and strengthen quality control through real-time monitoring and anomaly detection,” he adds.
“At Northeastern, our research at the intersection of the Bioanalytical Training Laboratory (BATL), the Center for Bioinnovation and Regulatory Sciences, and AI is exploring how AI can accelerate the development, manufacturing, and regulation of advanced therapies,” Auclair says.
Not plug-and-play
AI’s potential to spot patterns in data is attractive.
However, biopharmaceutical companies looking to adopt the technology are likely to encounter challenges, according to Auclair, who cautions that setting up an AI-driven manufacturing operation is about more than simply buying the right software.
“The technology is advancing rapidly, but successful implementation depends on having high-quality, well-curated data, digital manufacturing infrastructure, and multidisciplinary expertise spanning biology, engineering, data science, and regulatory science.
“AI is not a plug-and-play solution; organizations must build integrated data ecosystems and governance frameworks that regulators can trust,” Auclair says.
AI adoption is a multidisciplinary challenge and should involve people with expertise in all parts of drug development and production, according to study co-author Rominder Singh, PhD, professor of practice, regulatory sciences, & AI at Northeastern.
“Research conducted through the BATL and the Center for Bioinnovation, led by Professor Auclair, has focused on addressing many of these scientific and manufacturing challenges that are unique to advanced therapies.
“This is precisely why Northeastern’s pioneering work in RegSciAI is so important: it brings together regulatory science and AI to ensure these technologies are both innovative and deployable in real-world biomanufacturing,” Singh says.
In vivo CAR T sounds like the existing class of Chimeric Antigen Receptor (CAR) T-cell therapies used to provide often individualized treatment for cancer. But they’re a new and emerging class of therapeutics with their own challenges and opportunities for manufacturers.
That’s according to Mo Heidaran, PhD, chief scientist at Cellx, who is due to give a talk at the upcoming Bioprocessing Summit in Boston.
“Whether something is a cell or gene therapy, from a regulatory perspective, depends on [the nature of] the product that’s administered to the patient,” Heidaran explains.
“And, in the United States, in vivo CAR Ts are gene therapy products and not cell therapies as some people talk about them.”
The better-known CAR T products are ex vivo, delivered via modification of patient cells, he explains. Whereas this emerging class of therapies involves delivery of a genetically engineered virus or lipid nanoparticle (LNP) that, in some cases, is stably integrated into the patient genome.
According to Heidaran, the risk of integration is higher when viruses are used.
“My colleagues at the FDA want to make sure people understand it’s very important these products must be [designed] to be very specific to the cell type, perhaps based on data about [some of these] therapies having off-target effects,” he says.
Most in vivo CAR T-cell therapies are in very early stages, with none currently approved for patients, although Heidaran says they are increasingly under investigation by larger companies since they are scalable for a wider range of patients. Also, they are believed to be more cost-effective and have similar logistics, as they don’t require lymphodepletion, he adds.
“Essentially the value driver is that you’re pharmaceuticalizing cell and gene therapy since it’s just a vial of the virus or LNP that you can use to treat many patients—almost like a drug or pill,” he says.
Among the challenges for this emerging class is that several ex vivo CAR T-cell therapies are already approved for patients. In vivo CAR T therapies treat some of the same indications, i.e., certain cancers and autoimmune diseases, he says.
“At some point there has to be a decision made by the FDA about how these [new] therapies compare, such as [running] a study or external control as to whether they’re superior or non-inferior to the same or similar approved ex vivo CAR T,” he says.
Other challenges facing this new industry are about batch sizes for manufacturing, as the equipment and processes for treating ten patients are different from needing to treat thousands. Also, he says, in vivo CAR T therapies need to be monitored to look for off-target effects, durability of response, or an immune response by the patient.
“Overall, to develop a safety profile, we need to define what the effective dose is that people are working to, as these therapies may require repeat administration, which is not done with ex vivo-generated CAR T,” he says.
Researchers at Johns Hopkins University have developed a machine learning-based version of the widely used Martin-Hopkins equation that simplifies the calculation of low-density lipoprotein cholesterol (LDL-C) without compromising accuracy.
The new approach, which is published in JAMA Cardiology, could make it easier for laboratories to estimate LDL-C, improving treatment decisions for patients at risk of cardiovascular disease.
“We’ve optimized the calculation of LDL cholesterol and made this equation accessible and easier for all labs to implement,” said Seth Martin, MD, MHS, senior study author and director of the Advanced Lipid Disorders Program and Digital Health Lab at the Johns Hopkins Ciccarone Center for the Prevention of Cardiovascular Disease. “Our goal is to enable clinicians and patients to make better decisions about starting treatments that prevent heart attacks and strokes, and save lives.”
LDL-C is a major cause of atherosclerotic cardiovascular disease (ASCVD) and a primary treatment target. Current guidelines recommend using LDL-C cut offs, such as 70 mg/dL or 55 mg/dL (to convert to mmol/L multiply by 0.0259) in patients with ASCVD, to guide clinical lipid management.
The gold standard for measuring LDL-C concentration is preparative ultracentrifugation but this method is expensive and time-consuming. LDL-C concentrations are therefore usually estimated in routine practice.
One of the most accurate ways to estimate LDL-C concentration is the Martin-Hopkins method, which is recommended for clinical use in the U.S., Europe, and South America. However, implementation can be difficult because it requires users to look up an adjustable factor in a large table that is based on the patient’s triglyceride and non-high-density lipoprotein cholesterol levels.
“A lipid profile with low cholesterol and high triglycerides is the ultimate stress test of the LDL cholesterol calculation,” said Martin. He explains that a 5, 10 or 20 mg/dL difference, based on various equations, could change a person’s eligibility for treatment, such as with PCSK9 inhibitors, which have been shown to significantly lower LDL cholesterol levels. “It’s these types of on-the-cusp examples that benefit most from more accurate results,” he added.
To overcome this barrier and facilitate implementation, Martin and team used a transparent machine learning approach—multivariate adaptive regression splines—to create a simplified, formula-based LDL-C equation.
They trained and tested the tool on data from 4,939,528 adults and children (mean age, 56 years; 53% women) with complete lipid panel test results. These samples, which are representative of the U.S. population, had a median LDL cholesterol level of 114 mg/dL and came from the Very Large Database of Lipids.
The researchers report in JAMA Cardiology that the machine-learning version of the Martin-Hopkins equation estimated LDL-C concentrations that were similar to the original equation, with a minimal difference of 0.5 mg/dL.
Both Martin-Hopkins equations classified 90% of samples within the correct treatment category. Among other commonly used tools for LDL-C estimation, the Sampson-NIH equation correctly classified 86%, the modified Sampson-NIH equation classified 85%, and the Friedewald equation classified 83% in the correct category.
Importantly, said Martin, the investigators found that the Martin-Hopkins equations were the most accurate for classifying high-risk patients with lower ranges of LDL cholesterol levels.
When it came to assessing people who had triglyceride levels between 200 mg/dL and 399 mg/dL and LDL cholesterol levels less than 70 mg/dL, the Martin-Hopkins machine learning equation accurately classified 84% of high-risk samples, the original Martin-Hopkins equation classified 83%, the modified Sampson-NIH equation classified 72%, the Sampson-NIH equation classified 61%, and the Friedewald equation classified 40%.
Martin and co-authors conclude: “Given its high accuracy and straightforward implementation as a single line of code in laboratory information systems, [the Martin-Hopkins machine learning equation] is an alternative option to consider implementing in practice.”
Approximately three million people worldwide struggle with chronic pancreatitis, for which there is no cure. In a study published in Cell Stem Cell titled “Patient-derived organoids reveal ductal dysfunction and CFTR-modulator responses in chronic pancreatitis,” researchers from Salk Institute have developed an organoid platform to uncover the mechanism of chronic pancreatitis development and identify possible therapeutic strategies.
The authors generated 37 organoids from patients who developed chronic pancreatitis spontaneously. The organoids revealed consistent dysfunction in the protein cystic fibrosis transmembrane conductance regulator (CFTR), which was identified as a therapeutic target.
“Though patients can have the same clinical diagnosis of chronic pancreatitis, they can have very different underlying molecular drivers of that disease, which makes treatment especially difficult,” said Dannielle Engle, PhD, assistant professor at Salk and corresponding author of the study. “Our work breaks down a major barrier in the field by establishing an experimental model that preserves patient-specific disease biology and can be used to develop tailored therapies.”
Over the last decade, organoids have become a prevalent tool to bridge the gap between cell and human studies. Each organoid typically begins with stem or progenitor cells from patients. In Engle’s lab, donor pancreas tissues were used to create miniature replicas of the pancreas. Findings based on a patient’s personalized organoid model could improve therapeutic effectiveness.
“By growing organoids directly from patients, we preserve key features of ductal cells and ask which disease mechanisms are active in each individual patient,” said Victoria Osorio-Vasquez, PhD, a postdoctoral researcher in Engle’s lab and first author of the study.
The researchers surveyed the molecular signatures in each organoid and found three subtypes of chronic pancreatitis. This biology-based patient stratification can inform optimal treatment.Results showed that approximately half of the organoids demonstrated dysfunctional CFTR.
“And CFTR dysfunction was not limited to patients with inherited CFTR mutations, suggesting that functional testing may identify therapeutic opportunities that would be missed by genetic testing alone,” Osorio-Vasquez says.
Existing CFTR modulator therapies treat patients with cystic fibrosis. The findings suggest that these same therapies may offer pancreatic benefits. The researchers tested clinically available CFTR modulators and found that these therapies could stabilize or restore CFTR function and reduce inflammatory signaling in responsive pancreas organoids.
The platform also revealed rare alterations to genes, KRAS and TP53, in some chronic pancreatitis organoids, supporting future use of the system to study disease evolution, pancreatic cancer risk, and biomarker discovery at the interface of chronic inflammation and pancreatic cancer.
“These organoids gave us a way to study chronic pancreatitis pathogenesis in human cells for the first time,” says Engle. “Our platform enables a more personalized way of studying and eventually treating chronic pancreatitis, while also blazing the trail for other organoid-based platforms in other inflammatory disease contexts.”
This blog was originally posted by the TLC Foundation for BFRBs
Body-focused repetitive behaviors (BFRBs) and obsessive-compulsive disorder (OCD) are two distinct mental health conditions that share some similarities but also have significant differences. BFRBs involve repetitive, self-grooming behaviors that can cause physical damage, such as hair pulling or skin picking. On the other hand, OCD is a condition characterized by intrusive thoughts (obsessions) and repetitive behaviors (compulsions) performed to alleviate anxiety.
While both conditions involve repetitive behaviors and can impact daily life, their underlying mechanisms, triggers, and treatment approaches differ. This article explores the key similarities and differences between BFRBs and OCD to better understand these complex conditions.
Similarities Between BFRBs and OCD
Most professionals view BFRBs and OCD as similar conditions due to the similarity in symptoms, such as compulsivity and repetitive behaviors. These two conditions share several similar systems and are usually a reaction to triggering factors such as stress and anxiety. Below are some of their similarities.
Repetitive Behaviors
Individuals dealing with BFRBs often engage in various repetitive behaviors such as hair pulling, lip biting, or skin picking. These actions are usually challenging to control and are frequently triggered by stress or anxiety. One may indulge in the habit subconsciously to find instant relief from the trigger.
Individuals with OCD often experience intrusive thoughts that result in repetitive behaviors known as compulsions. Some common compulsions include washing hands and repetitively checking or counting to alleviate the stress caused by obsessive thoughts. In both conditions, the repetitive behaviors are often exacerbated by stress and anxiety, and individuals may adapt these behaviors as a coping mechanism.
Impulse Control
Closely related to repetitive behaviors is the concept of impulse control. Both BFRBs and OCD involve challenges in this area, albeit in different ways. Individuals with BFRBs and OCD may find it hard to control the urge to perform repetitive behaviors. This is because these repetitive behaviors often relieve tension. Despite knowing the consequences of these behaviors, the desire to indulge in them is usually irresistible.
For example, individuals with BFRBs understand that hair pulling may affect their appearance, but they cannot refrain from doing it. OCD occurs as a result of intrusive thoughts whereby one believes that if they do not perform a specific action, the stressor won’t go away. These intrusive thoughts often cause anxiety, which can be eased by engaging in the said repetitive behavior.
Onset and Course
Having examined the behavioral aspects, let’s now consider how these conditions develop over time. The onset of these two conditions shares several similarities regarding age, triggers, and psychological mechanisms.
The onset of both conditions is usually during childhood or adolescence and often coincides with various developmental changes and stressors. For individuals with BFRBs, the repetitive behaviors alleviate stress and anxiety instantly. At the same time, for those with OCD, performing the compulsions temporarily relieves them from the stress caused by their intrusive thoughts. The cognitive patterns involve repetitive actions, intrusive thoughts, and a lack of impulse control. In BFRBs, the urge to engage in these repetitive behaviors can be intrusive and persistent, while in OCD, one’s obsessions create a sense of urgency, which leads to the adoption of compulsive actions.
Neurobiological Factors
To fully understand the similarities between BFRBs and OCD, we must delve deeper into their biological underpinnings. Both conditions have a genetic origin and are associated with neurobiological factors. Neurobiological studies indicate that the impulse control and emotional regulation difficulties for people with BFRBs and OCD are often caused by abnormalities in brain regions that are responsible for impulse control and habit formation. Therefore, the underlying brain mechanism may result in the onset and development of both conditions. It is not uncommon for individuals to have both BFRBs and OCD or for both conditions to coincide with other mental health conditions, usually depression and anxiety. The overlap is generally because they typically share common underlying factors that play a part in their severity and development.
Differences Between OCD and BFRBs
While BFRBs and OCD share several commonalities, it’s equally important to understand their distinct characteristics, from the symptoms to the underlying mechanisms. Let’s explore the key differences that distinguish these two conditions.
Nature of the Behavior
First and foremost, let’s examine how the behaviors associated with each condition differ in their fundamental nature. Individuals dealing with these two conditions adopt diverse behaviors as coping mechanisms for their triggers. In BFRBs, the behaviors adopted, such as trichotillomania (hair-pulling) or cheek-biting, usually result in physical harm. However, regardless of the consequences, one always feels relieved when picking their skin or pulling their hair.
OCD, on the other hand, involves a wide range of compulsions, from washing to organizing, checking, and counting. Compulsive behaviors are performed due to intrusive thoughts that make one think that if they fail to indulge in a specific behavior, they might get hurt, or there might be other negative consequences.
Presence of Obsessions
Another crucial distinction lies in the cognitive processes behind these behaviors. Generally, BFRBs do not involve obsessive thoughts. The primary focus on BFRBs is usually more on the physical behavior and not the fear of specific consequences.
However, the major characteristic of OCD is intrusive thoughts, which increase the urge to indulge in particular behaviors for relief. The thoughts are usually persistent with unwanted images that result in distress.
People with BFRBs DO NOT report that if they do not pick on their skin, something terrible will happen. Instead, they report that picking or pulling their hair helps relieve them from intense and negative emotions. These behaviors, therefore, serve a self-regulatory function, unlike in OCD, where the repetitive behavior calms them from their intrusive thoughts.
Triggers
The nature of triggers for each condition is closely related to the presence or absence of obsessions. The primary trigger in BFRBs is stress and anxiety, but for OCD, the main trigger is intrusive thoughts, which then result in anxiety. OCD and BFRBs triggers differ in several ways, often resulting in different outcomes. OCD triggers often result in one taking measures to prevent harm, while for BFRBs, one uses the adopted behaviors to regulate and manage intense emotions. The nature of thoughts is an essential distinguishing factor, seeing as OCD involves intrusive and obsessive thoughts that trigger specific behaviors adopted to prevent harm. The purpose of compulsions in OCD is to reduce the anxiety caused by the obsessive thoughts, while in BFRBs, the behaviors are for emotional relief.
Awareness
Beyond triggers, the level of conscious awareness also differentiates these two conditions. Those dealing with BFRBs usually find themselves biting their nails or even pulling their hair subconsciously. Individuals with OCD are generally aware of their intrusive thoughts and are compelled to adopt specific behaviors as a response to these thoughts. Individuals with OCD are often aware of their compulsions and understand when they are being irrational, but they are unable to control themselves. Compared to people with OCD, those with BFRBs often find their behaviors more rewarding than distressing.
Treatment
Finally, while both conditions may benefit from cognitive behavioral therapy, the specific approach to treatment varies significantly. For individuals with BFRBs, the focus is on behavior modification and awareness, achieved through habit reversal training. For OCD, the emphasis is often placed on exposure to anxiety-provoking thoughts to help an individual tolerate anxiety, which prevents compulsive behavior.
Bottom Line
While BFRBs and OCD can coexist, they are distinct disorders with unique manifestations despite sharing some similarities. The key distinctions between these conditions are evident in their underlying mechanisms and treatment approaches.
Both involve compulsive behaviors, but their purposes differ. BFRBs primarily serve as subconscious tools for emotional regulation. OCD compulsions are conscious attempts to alleviate anxiety and prevent perceived harmful consequences. BFRB behaviors often occur with limited conscious awareness, while OCD sufferers are typically more aware of their compulsive actions.
Both conditions can significantly affect daily functioning and social interactions.BFRBs may lead to physical injuries and lowered self-esteem due to visible effects. OCD can cause severe anxiety and time-consuming rituals that interfere with daily activities.BFRB treatment emphasizes behavior modification and awareness techniques, while OCD treatment often involves exposure therapy to reduce anxiety responses.
Understanding these distinctions is crucial for accurate diagnosis and effective treatment. While both conditions present challenges, with proper support and intervention, individuals with BFRBs or OCD can learn to manage their symptoms and improve their overall quality of life.
<![CDATA[FDA accepts AXS-12 NDA for cataplexy in narcolepsy after phase 3 shows fewer attacks, better wakefulness, and cognition; decision due May 2027.]]>
For decades, psychiatry has used trial-and-error symptom-based diagnoses and treatments. Standardized diagnostic frameworks brought much-needed consistency to the field, but they also grouped diverse patients with different biology. Consequently, many people receive unsuitable treatments.
On this episode of Behind the Breakthroughs, Alto Neuroscience founder and CEO Amit Etkin, MD, PhD, discusses how precision medicine will change mental health care. Etkin explains how objective biological measures like cognitive testing, EEG brain activity, sleep and circadian rhythm monitoring, and advanced computational analysis can help identify patients who will benefit from specific therapies rather than just symptoms. Comparing psychiatry to precision oncology, he explains why it is at a turning point. Instead of finding a perfect biomarker, the field is developing practical, scalable tools to link brain function to targeted drug development.
Etkin shows from Alto’s clinical pipeline how matching therapies to biologically defined patient populations can improve outcomes and reduce psychiatric treatment uncertainty. We also examine the potential and limitations of genetics, multi-omics, wearables, and AI in precision psychiatry. Etkin explains why brain measurements may be more clinically useful than peripheral biomarkers and how AI can help find patterns in complex biological data. Finally, we discuss how precision psychiatry will become routine clinical practice, from regulatory acceptance and standardized data collection to the first biomarker-guided therapies. If successful, these advances could transform psychiatric disorder diagnosis, treatment, and understanding.
This interview has been edited for length and clarity.
IPM: What has been the key limiting factor to advancing precision psychiatry?
Etkin: There was a period of time in the 1940s, 1950s, and 1960s when psychiatry was beginning to develop. Our definition of diseases was bespoke to how you practiced them. They really made very little sense.
Amit Etkin, MD, PhD, co-founder and CEO of Alto Neuroscience [Alto Neuroscience]
The DSM did a great job of bringing everybody under the same diagnostic umbrella. We can talk about a common set of symptoms and everybody’s talking about them. Yes, we have wonky definitions of diseases, but at least we’re starting to use the same definition.
We are finding large, heterogeneous groups of people that fit into a category. If you examine the words used to describe some psychiatric labels from the past, they sound antiquated because they are linked to outdated concepts and are inconsistently defined. That was good. The problem is that we never transitioned beyond that.
I try to focus on a much simpler approach to what we’re doing. Instead of thinking about biomarkers, machine learning, and so forth, it’s about knowing what we are doing. If what you’re doing is developing a drug for a population with depression, it’s a massive category where some people just symptomatically might sleep more and others sleep less, and some eat more and others eat less. There seems like it’s a bit of a mess. What would you want to know? At a very basic level, that would simply allow you to know what you’re doing better. What would you like to measure?
There are any number of answers. We know that some patients with depression have cognitive problems and others don’t and that people with cognitive problems have worse outcomes regarding standards of care and treatment. We’ve known that for a long time. But why aren’t we doing that measurement systematically in every single drug trial or in clinical care? Because if we did that one little thing systematically, we would see that some treatments or people saw it one way and others saw it another way just by collecting that data more systematically. That’s what we had been doing in the lab: figuring out what kind of data to collect systematically.
That’s what led us to the form Alto, understanding that there are certain measures, like cognition and brain activity. We do this non-invasively with electroencephalogram (EEG) brainwave recordings and wearables to look at circadian rhythms that are physiologically and biologically meaningful, easily scalable, low cost to meet our health system’s needs, and, if done consistently, insightful for separating populations and identifying mechanisms to develop for them and for translating back to an animal model, which is not possible if you only look at depressive symptoms.
Do we know what we’re doing now? No. We’re really just beginning on this journey as a field. But I do think people now recognize that this is the direction of travel. You can see that more and more companies and academic research centers are emerging under this theme.
IPM: What are your thoughts on the use of genomics, multi-omics, or blood-based biomarkers of the central nervous system?
Etkin: The way I interpret that literature is that there’s smoke but not yet fire. The classical molecular marker is genetics. We get that on every patient, but we do not use stratification here, as genetics cannot achieve meaningful stratification in our populations. It’s predominantly common variance with each of them, or in combination, having a very rare large effect size variance, which is really not going to be clinically meaningful from a drug perspective because you’re treating one out of every 10,000 people or whatever the prevalence is.
I don’t know the best polygenic risk score for schizophrenia, determined from 100,000 people, which is probably the high watermark for psychiatric genetics and may explain 1–2% of the variance. You needed to explain at least 10% to stratify the population even a little bit. I don’t think genetics will ever get us there for that purpose.
Where there’s smoke but not yet fire are immune measures. There have been many implications for different immune measures in psychiatric disorders, but every time it gets tested to see whether people with high inflammation respond to something that targets our process directly or indirectly, those studies never really work out.
Then there’s the even larger world of multi-omics, where you have a multiple testing issue and, fundamentally, the problem that what you’re sampling is very peripheral to the organ that matters. Where is the serotonin in the blood coming from? It’s mainly coming from platelets—it’s not coming from the brain. The bulk of serotonin in the body is in the gut. So you can measure… a ton of different proteins, different configurations, and modifications to those proteins probably have relatively little purchase on what is going on in the brain.
The simplest example is a protein called brain-derived neurotrophic factor (BDNF), which is a really important neuroplasticity protein in the brain. It’s also found in the blood, and people tried over and over and over again. You see some positive studies, but mainly studies that are negative and some that are just not published that come to the conclusion that there’s very little bearing of what you’re measuring peripherally to what’s going on centrally.
Measuring the brain directly with EEG or the output of specific brain circuits through behavioral tests is much more amenable and has better performance statistics and interpretability for gaining insights.
IPM: Has a specific layer or test modality enabled precision psychiatry programs for Alto Neuroscience, or are they ultimately based on aggregate measures?
Etkin: Less so in aggregate as measured together but each alone. We try not to combine everything into one model because it becomes very complicated, and we have already been told by the FDA in no uncertain terms that a multimodal biomarker is not something we will readily consider. To get a multimodal biomarker approved for some sort of use, you have to validate each and every component alone and their combination, which sounds like a headache. But I’m not sure you necessarily need to either.
What I would consider to be a win is getting a drug for the whole population with a marker that enriches finding ways to show additional value in a drug program through a biomarker perspective and, over time, an iteration. The field then transforms into one where oncology already exists, which means they expect you to know what you’re doing. You have to understand the population. You have to understand how your drug impacts the biology that defines the population. We’re not there yet.
But they weren’t there yet either, in the same kind of single stroke that we now envision. It was like the first precision therapeutics, like Herceptin, were approved well over a decade before the immuno-oncology (IO) revolution. That really brought precision oncology into maturity, as we understand it now. That history suggests we probably need our IO moment as an inflection point, but we are not yet ready for it.
We need that Herceptin moment first: start transitioning how people think and collect data and create a bit more of a common language across programs so that different drug makers and different academic labs aren’t collecting their own unique data sets that aren’t then harmonized across them. You can’t speak about a thing as an invariant measure of a process that doesn’t matter who is measuring; they get the same outcome.
The field has been somewhat resistant, probably for cultural reasons related to how people have historically operated, to a lot of data sharing and harmonizing of what we’re collecting, how we’re collecting it, and how we’re analyzing it. We’re just starting to really move in that direction. Those are all limits that gate the early-stage biomarker collection efforts.
IPM: Is there a future where someone walks in with a psychiatric condition and undergoes a battery of measurements that spits out a drug that has a high probability of being effective?
Etkin: I think that bar is a lot lower than that. It doesn’t have to be very effective—it just has to be more effective than chance. because that’s where we are. If I told you that instead of a 30% chance of remission with a drug, I could increase it to 40% or 45%, would that be helpful? That’s meaningful. You convert that to a number needed to treat it. For a clinician, this represents a significant effect, although it does not achieve perfect precision. All you need is something better than nothing, which is what we have. Of course, once you have something that’s better than nothing, now you have a new benchmark, and things will continue to improve, which is great. But the field’s got to start somewhere.
I don’t think it is that far away. I think in our efforts and the efforts of others in the field who have followed suit and taken a precision approach, something will work. When that changes, all of a sudden you can’t envision going back; it’s only going forward. That’ll be super exciting. It’s not like a “by the time I retire” kind of thing. It’s within the next five or a maximum of ten years that we will be at that inflection point.
IPM: Does precision psychiatry apply to the rest of neurology?
Etkin: If you anchored on the way I framed brain circuit function earlier. And what’s measurable is that there is no line between psychiatry and neurology. You have neurologists who are called “functional neurologists” or something in that vein, where they think about what I would call the “psychiatric aspects of neurology.”
A big part of Parkinson’s is cognitive impairment in a substantial portion of people, leading to dementia. Nothing to do with the movement disorder, but everything to do with the biology affecting different circuits. The right mood, in fact, is one of the earliest areas of perturbation in Parkinson’s that will then predict the development of the motor symptoms. Some people have perfectly well-controlled motor problems but have cognitive problems and mood problems that are even more prominent and lead to more, especially on the cognitive side, of their ultimate disabilities. Cognitive impairments are even a contraindication for deep brain stimulation.
Because of these interactions, all of these boundaries are artificial. It’s just a core engineering question of, can I know what I am measuring and what I am manipulating? There’s no reason we need to draw that line in an artificial way. It’s just about whether I can leverage the tools and the drugs in a useful way together.
