University Hospitals Birmingham NHS Foundation Trust and Miziac will digitise more than 235m patient record images by December 2026.
Mental Health Fitness Through a Youth Perspective
Insights on building and advancing mental health care solutions through collaboration — from the Global Youth Advisory Council at the SNF Global Center for Child and Adolescent Mental Health at the Child Mind Institute
Mai El Shoush, Partnerships Campaign Manager, Stavros Niarchos Foundation (SNF) Global Center for Child and Adolescent Mental Health at the Child Mind Institute
Tatiana P. Claridad, MBA, Director of Board Affairs and Institutional Strategy, Stavros Niarchos Foundation (SNF) Global Center for Child and Adolescent Mental Health at the Child Mind Institute
Mental health fitness is shaped by lived experience, context, and the systems of care that surround us. During Mental Health Awareness Month, we invited young leaders from the Global Youth Advisory Council (GYAC) at the Stavros Niarchos Foundation (SNF) Global Center at the Child Mind Institute — from Brazil, South Africa, and Greece — to share their unique perspectives.
Their reflections offer insight into how young people define, build, and experience mental fitness globally, highlighting the influence of their environments, communities and everyday realities. Their views are essential to shaping more relevant and effective support for young people everywhere.
From daily habits to systemic change, here’s how young people are reimagining mental health care as mental health fitness:
These reflections raise a broader question — how can global collaboration strengthen mental health support for children and adolescents?
While these insights from the GYAC members highlight the importance of collaboration, they also reflect a new paradigm of youth leadership.
What continues to inspire you as a young leader about the ideas and perspectives shared through the Global Youth Advisory Council, and what does it say about the future of mental health care?
“What continues to inspire me most is the diversity of perspectives across different countries, yet the shared commitment to improving youth mental health. There is something powerful about young people coming together across contexts, bridging gaps through a global lens while staying rooted in their lived experiences. It reminds me that the future of mental health care will be more inclusive, shaped by real voices, and focused on breaking stigma in ways that feel authentic and meaningful.”
Delice Lumbu, 20, Director of Youth Engagement, SNF Global Center
Contributors: Delice Lumbu, Director of Youth Engagement, Stavros Niarchos Foundation (SNF) Global Center for Child and Adolescent Mental Health at the Child Mind Institute
The post Mental Health Fitness Through a Youth Perspective appeared first on Child Mind Institute.
Derbyshire Healthcare to roll out Psyomics mental health platform
Derbyshire Healthcare NHS Foundation Trust will roll out Psyomics’digital platform to improve access to community mental health care.
The Sentinel Phenotype: a theoretical bioenergetic and neurobiological framework for high-fidelity predictive systems (HEPOE Theory)
The HEPOE Theory (High Entropy Predictive Organization Efficiency) proposes a novel conceptual framework for understanding Giftedness (HA/G), moving beyond academic performance-based models toward a biophysical and neuroscientific foundation. Through a theoretical synthesis grounded in the Free Energy Principle and Biological Thermodynamics, the gifted individual is redefined as a “Sentinel”: a high-fidelity sampling system specialized in the early detection of isomorphy and the reduction of systemic entropy. This framework reinterprets Charles Spearman’s general intelligence (g) as a macroscopic manifestation of hardware efficiency, where reasoning ability is proposed to be fundamentally constrained by working-memory capacity and the metabolic economy of ATP resynthesis. We hypothesize that the hardware operates under an “open sensory gating” regime and low latent inhibition, leading to high metabolic costs and chronic allostatic load. The paper introduces the original concept of Predictive Moral Injury to conceptualize the potential somatic damage resulting from the early perception of ethical-systemic collapses within low-resolution environments. The HEPOE Unification Matrix integrates decades of classical literature and proposes a rigorous differential diagnosis against the pathologization of ASD, ADHD, and PTSD. It hypothesizes that the Sentinel’s exhaustion is not a dysfunction, but a logistical byproduct of high predictive performance under entropy-saturated conditions.
Negative Online Experiences Are Common but Often Go Unreported Among Youth With Mental Health and Neurodevelopmental Concern
New Child Mind Institute study finds more than one in four youth experienced a negative online experience in the past year, yet only one in five reported the incident through platform tools.
New York, NY — A new study from researchers at the Child Mind Institute finds that negative online experiences are common among children and adolescents with mental health and neurodevelopmental conditions, and that most incidents are not reported through platform reporting tools.
Published in JAACAP Open, the study examined negative online experiences among 1,009 youth ages 9 to 15 with a history of mental health or neurodevelopmental concerns, all of whom were current or previous participants in the Child Mind Institute’s Healthy Brain Network. More than one in four reported at least one negative online experience in the past year. Among those who had such an experience, nearly 69% reported multiple incidents, yet only 20% reported the incident through platform reporting tools.
The study defined “negative online experience” as any unwanted or uncomfortable experience while online, including cyberbullying, cyberstalking, doxxing, impersonation, sexual harassment, and related forms of digital harm. The research used a mixed-methods design, combining a quantitative survey with an in-depth qualitative follow-up involving a three-day moderated online bulletin board with a subset of participants.
“These findings point to a large and often hidden problem,” said Michael P. Milham, MD, PhD, Chief Science Officer at the Child Mind Institute and senior author of the study. “Many young people are encountering harmful or uncomfortable experiences online, but the systems designed to help them often do not receive a report. That creates a major gap for parents, educators, clinicians, and platforms trying to keep children safer online.”
The research team identified three major categories of barriers that prevent youth from reporting negative online experiences: reporting process barriers, such as not knowing how to make a report; reporting policy barriers, including uncertainty about what qualifies for reporting or how platform rules apply; and emotional barriers, such as embarrassment, fear, and worry about consequences.
The study also found that reporting decisions were often shaped by how young people interpreted the incident itself. In the qualitative follow-up, youth considered whether the harmful behavior seemed intentional, how malicious it appeared, and how severe or repeated the harassment was. When those cues were ambiguous, youth were less certain about whether reporting was appropriate.
“Reporting is not simply a matter of telling young people to speak up,” said Mirelle Kass, lead author of the study. “Youth are making complicated judgments about intent, severity, platform rules, and the possible consequences of disclosure. If we want young people to report harmful experiences, the tools and systems around them need to be clearer, safer, and easier to use.”
The findings suggest that online safety efforts should be tailored to the needs of youth who may already be managing mental health, developmental, or social challenges. Social aptitude, mental health symptoms, and parenting style were associated with youths’ likelihood of encountering negative online experiences and with the barriers they faced when deciding whether to report them.
Participants also expressed a clear desire for better tools and guidance. Most youth wanted platforms to provide more information about how to protect themselves online, how to use safety features such as blocking and reporting, and how to access support during and after the reporting process.
“Families, educators, clinicians, policymakers, and technology developers all have a role to play,” said Dr. Milham. “We need reporting systems that children can understand, policies that are transparent, and trusted adults who can respond without blame or overreaction. Safer digital spaces will require more than awareness. They will require systems designed around how young people actually experience online harm.”
The study underscores the importance of developmentally appropriate safety tools, clearer platform policies, and stronger support systems for youth navigating digital spaces. For children and adolescents with mental health and neurodevelopmental conditions, improving reporting pathways may be an important step toward reducing hidden online harms and building safer online environments.
This research was supported by funding from Google LLC’s User Safety team to the Child Mind Institute for work led by Michael P. Milham, MD, PhD.
About the Healthy Brain Network
The Healthy Brain Network is a community-centered research initiative from the Child Mind Institute that collects clinical, cognitive, behavioral, and neurobiological data from children and adolescents in the New York City area. Families who participate receive feedback and diagnostic consultation while contributing to open science research aimed at improving understanding of child and adolescent mental health.
About the Child Mind Institute
The Child Mind Institute is an independent nonprofit organization dedicated to transforming the lives of children and families struggling with mental health and learning disorders. Through cutting-edge research, evidence-based clinical care, and public education, the Child Mind Institute builds open science platforms and digital tools to accelerate discovery and improve youth mental health worldwide.
For press questions, contact cmiscience@ssmandl.com or mediaoffice@childmind.org.
The post Negative Online Experiences Are Common but Often Go Unreported Among Youth With Mental Health and Neurodevelopmental Concern appeared first on Child Mind Institute.
Recruitment and Participation of Black Home Health Care Patients in Speech-Based Cognitive Research: Mixed Methods Feasibility Study
Background: Older Black adults remain underrepresented in dementia research, particularly in studies using speech-based methods for early cognitive assessment. Understanding how to effectively recruit and engage this population in research involving audio-recorded interactions is critical to ensuring equitable inclusion and developing culturally responsive study designs. However, recruiting older Black adults into dementia research remains a significant challenge. Objective: This study assessed the feasibility of recruiting older Black home health care (HHC) patients into speech-based cognitive research and examined the factors influencing participation and participants’ data collection experiences. Methods: We conducted a convergent mixed methods feasibility study using a 4-component recruitment pipeline at a nonprofit HHC agency: (1) patient identification and study introduction, (2) in-home audio-recorded cognitive assessments, (3) follow-up calls, and (4) audio-recorded patient-clinician encounters. Both patients and their corresponding clinicians were included in this study. Eligible participants were Black adults aged 60 years and older receiving HHC services in New York City. Patient demographic and clinical characteristics were compared between those who consented and those who declined using bivariate analysis. Qualitative feedback was gathered through patient questionnaires and clinician semistructured interviews and was analyzed using reflexive thematic analysis. Results: Of 246 patients contacted, 71 (28.9%) provided verbal consent and 60 (24.4%) completed cognitive assessments. Five patients were excluded due to health conditions or severe cognitive impairment, leaving 55 eligible participants. Among these participants, retention remained high across study components, including follow-up calls (48/55, 87.3%) and audio-recorded clinician visits (54/55, 98.2%). Patients who did not consent were more likely to have greater complex medical profiles, including higher pain interference (=.01), need for assistance with medication reading (=.04), and polypharmacy (≥5 medications; =.01), while no significant differences were observed for age, gender, or functional status. Qualitative findings demonstrated high acceptability and feasibility of audio recording. Patients reported strong motivation to participate, positive and engaging experiences, comfort with recording, and minimal disruption to care. Clinicians reported ease of integration into workflow, initial discomfort that diminished over time, minor technical challenges, and perceived benefits for communication and patient engagement. Conclusions: Recruiting Black older adults receiving HHC into speech-based dementia research was feasible and well accepted. Culturally tailored recruitment may enhance equitable participation and guide future research using audio-recorded speech for early cognitive detection.
<img src="https://jmir-production.s3.us-east-2.amazonaws.com/thumbs/b5818e122d1c9d95d1dad3f3e7fea9cc" />
Mechanisms of action and advances in application of music therapy in improving cognitive impairment based on the theory of neuroplasticity
As global population aging accelerates, preventing and treating cognitive impairment has become a major public health priority. Music therapy is a safe, well-tolerated non-pharmacological intervention with substantial potential to improve cognitive function. This review synthesizes the neurologic music therapy (NMT) framework, encompassing techniques targeting attention, memory, and executive function, delivered through both active and receptive approaches. Clinical investigations indicate that music therapy may improve cognition and neuropsychiatric symptoms in Alzheimer’s disease (AD), vascular cognitive impairment (VCI), Parkinson’s disease–related cognitive impairment, mild cognitive impairment (MCI), and traumatic brain injury (TBI); however, effects appear to vary by intervention duration and disease stage. This narrative review aims to provide a theoretical foundation and practical guidance for the non-pharmacological intervention of cognitive impairment by collating evidence on the neuroplasticity theoretical foundations, technical systems, and clinical applications of music therapy.
Centering People, Centering Stories: Folklore as an Unlikely Ally in the OCD Misdiagnosis Crisis
By: Rebecca Bernstein, MA Folklore
The International OCD Foundation’s (IOCDF) recent landmark white paper reveals more than 80% of OCD cases in America remain undiagnosed (International OCD Foundation, 2025). Considering the size of this clinical challenge, it might seem odd to suggest that a small, humanities-based field like folklore— yes, folklore — has any role to play in the solution. As someone who studies OCD personal narratives (stories people tell about their lived experiences), my research suggests otherwise. In a situation that invokes the feeling of all-hands-on-deck, the tools and perspectives of this field may offer more benefit than we might initially give it credit for.
Folklore is the study of informal, creative communication. Dr. Lynn McNeill describes it as anything people “say, do, make, or believe” (McNeill, 2013). Folklore includes everything you’d think of (quilts, traditional music, fairy tales) and a lot of things you wouldn’t (occupational culture, gossip, internet memes.) We find examples of folklore everywhere. It’s in our holidays and our hobbies, our food and our fads, our jokes and our grieving. Folklorists study the infinite ways people express themselves in daily life. This, in turn, helps us better understand the cultural realities in which they live. And because what we “say, do, make, or believe” describes most of human behavior, the folkloric lens can be an indispensable one with which to investigate the world.
The benefit of studying how people express themselves is obvious when we recognize that in mental health, conversation and narrative are the primary tools we use to give and receive care. OCD isn’t just a diagnosis. It’s also a fundamentally creative experience. (Creative things don’t necessarily have to be beautiful, pleasing, or even wanted. They just have to be new and meaningful.) (Dictionary.com, 2023). Although ego-dystonic, extraordinary beliefs and elaborate rituals are hallmark features of OCD. When sufferers discuss their experiences, they are naturally inclined to do so through their own personal and cultural lenses. Therefore, descriptions of OCD vary infinitely. If the issue is our failure to recognize OCD when it presents itself, an approach designed to make sense of something as messy as human expression may offer insights that quantitative research methods still struggle to obtain.
How Folklorists Research
Just like in biomedical research, the research methods folklorists use matter. Our goal is to better understand people and their communities. That means we strategically build relationships, listen deeply, and intentionally embrace the complexity of those we talk to.
When I started researching OCD narratives, I wanted to know: What were the internal realities like for people who lived with this illness? What made their stories distinct? And how might those stories be connected? One of the biggest challenges I faced in my fieldwork was the potential for my participants to self-censor. As someone who also lives with OCD, I knew all too well the role shame and fear could play in the choice to fully share one’s reality with others. Using both field-tested approaches and my own lived knowledge, I conducted interviews with people with OCD, approaching them in a way I hoped would ease interviewees into difficult conversations:
- I provided anonymity. I held all interviews on Zoom, where participants were free to keep their cameras off. I also assigned each one an alphanumeric signifier (A1, B2, etc.) in my writing.
- I emphasized the importance of story. Although I asked specific questions, I also allowed participants to go off topic and engage in two-way conversation. The story was the point.
- I used the “kitchen table” interview method. Based on the work of Carl Lindahl, this method tries to recreate the intimacy of two individuals talking around a kitchen table. It discourages framing the interviewer as an objective party, recognizes storytellers as experts in their own experiences, and suggests that interviewers only ask questions they themselves would be willing to answer (Lindahl, 2012).
- I disclosed. My choice to openly discuss my own OCD diagnosis with interviewees allowed conversations to proceed with a certain warmth and vulnerability.
- I emphasized participants’ humanity. I treated each participant as a full individual rather than just a source of information. This meant I worked on a model of enthusiastic consent. It also meant I asked them for feedback on my writing to ensure I portrayed their experiences accurately.
- I compensated participants well. Each received a $100 gift card.
The Results
The universal theme I discovered during these interviews was a profound concern with social isolation. Every single participant mentioned this issue. Interviewees shared how OCD made it difficult for them to maintain relationships and how challenging it was to hide their illness from others. They also recalled their joy and gratitude when discussing moments in which they felt understood.
Their narratives also contained four other common themes:
1.) Logic and patterns of personal concern. Participants often discussed their particular obsessions and compulsions, and the influence those specific thoughts and behaviors had on their daily lives.
2.) Issues of negotiation. People talked about navigating certain types of conflicts as a result of their illness. These conflicts generally fell into two categories: self-negotiation and existential negotiation. In the first, people struggled with the desire to take their thoughts and urges seriously despite knowing they didn’t make sense. In the second, they wrestled with their relationships to the divine.
3.) Positive approaches to the illness. Many interviewees made a point to mention the silver linings they saw in being sick. They noted how OCD made them safer, more empathetic, or provided them with particular skills. Others discussed productive choices they’d made despite living with such a debilitating condition.
4.) Interactions with medical systems. Participants talked about their experiences as patients. For some, dealing with doctors, therapists, and other health professionals helped them understand their experience or relieved their suffering. For others, these encounters were confusing, unhelpful, or even traumatizing.
It’s important to note these themes represent a truly broad range of content. Not every story included every theme, and within those themes, the specific details I heard varied as much as the individuals themselves.
Implications
Say you were to hear four stories: one about someone’s preferred cleaning routine, one about someone’s waning belief in God, one about a good decision made in a difficult circumstance, and one about a doctor’s visit. It’s unlikely you’d consider these stories connected. And yet the data shows they are. The fact that stories with dramatically different content can reflect the same illness highlights the way OCD can remain elusive and camouflaged.
The problem with recognizing these stories as OCD stories isn’t just the variation in content. It’s also in how others hear them. In folklore, we don’t just study cultural expressions. We also study how they move from person to person. “Tellable narratives” travel easily. Both speakers and listeners understand what a certain type of story should sound like and the meaning it’s supposed to convey. If I tell you a tale about a persecuted young woman who escapes a bad home life and marries a prince, you can probably guess you’ve heard Cinderella. If we’re both excited that she went from rags to riches, we share an understanding that her journey is a positive one. In contrast, an “untellable narrative” hits some kind of barrier. If you’ve never heard Cinderella before or think the stepmother is actually the hero, my meaning in telling you the story gets lost. Untellable narratives can be misinterpreted.
This misalignment between the stories people tell and the ones listeners expect to hear happens all the time. We’ve all said things misunderstood by others. Sometimes this process is harmless; other times it results in difficult consequences. Dr. Kristiana Willsey writes about veterans who censor themselves in front of civilian audiences. Because civilians usually only expect to hear tales of “war heroes” or “PTSD survivors,” veterans often choose not to tell the full and complicated stories of their service experiences (Willsey, 2015). Dr. Amy Shuman and Carol Bohmer discuss the case of rejected asylum seekers. If asylum applicants don’t tell their stories of oppression and escape in a way that fits immigration officials’ expectations of what a traumatic asylum story should look like, their applications get denied (Shuman & Bohmer, 2016). If we consider just how different any two OCD stories can be and add the public assumption that OCD is an illness of specific doings (hand washing, checking locks) rather than tellings, it highlights just how difficult it is for most of these narratives to get heard, and heard correctly.
Patient/practitioner interactions can be particularly vulnerable to this type of miscommunication. The problem with considering OCD as just a medical issue is that most people don’t think of their lives as medical events. Practitioners enter the room ready to make sense of problems in clinical terms. Patients enter with stories. They share their concerns in a way that cannot be easily separated from their personal frames of reference or cultural understandings of life. Practitioners are often taught to mistrust the details that emerge from these narratives, to kindly but efficiently work around them in order to do their jobs. But for patients, these details are how they make meaning. If misdiagnoses also occur during these interactions, it’s worth taking a closer look at what’s being lost in translation.
Folklore ultimately offers the promise of new solutions to old problems. It allows us to reconsider how we listen to patients, collect data, and address communication issues— all clear benefits in the fight for better diagnostic care. It is also equipped to help us make sense out of the lived reality of OCD— perhaps uniquely so. I see folklore as an exciting potential ally to traditional research and clinical spaces. My hope is that this partnership can help us work more effectively toward our common goals: a better understanding of OCD, and quicker ease for its sufferers.
Works Cited
Dictionary.com. (2023). Creativity. In Random House Unabridged Dictionary. Random House, Inc. https://www.dictionary.com/browse/creativity.
International OCD Foundation. (2025). America’s OCD care crisis: National findings on the failure of effective OCD treatment to research patients. International OCD Foundation. https://iocdf.org/wp-content/uploads/2025/12/Full-Report-Americas-OCD-Care-Crisis-12-9-2025.pdf.
Lindahl, C. (2012). Legends of Hurricane Katrina: The right to be wrong, survivor-to- survivor storytelling, and healing. The Journal of American Folklore, 125 (496), 139–176. https://doi.org/10.5406/jamerfolk.125.496.0139.
McNeill, L. (2013). Folklore rules: A fun, quick, and useful introduction to the field of academic folklore studies. Utah State University Press. https://muse.jhu.edu/book/27822.
Shuman, A. & Bohmer, C. (2016). The stigmatized vernacular: Political asylum and the politics of visibility/recognition. In D. Goldstein & A. Shuman (Eds.), The stigmatized vernacular: Where reflexivity meets untellability. Indiana University Press.
Willsey, K. (2015). Falling out of performance: Pragmatic breakdown in veterans’ storytelling. In T.J. Blank & A. Kitta (Eds.), Diagnosing folklore: Perspectives on disability, health and trauma. University Press of Mississippi.
The post Centering People, Centering Stories: Folklore as an Unlikely Ally in the OCD Misdiagnosis Crisis appeared first on International OCD Foundation.
Precision, Gene-Edited Antimicrobial Targets Difficult-to-Treat E. Coli
An oral, next-generation antimicrobial created using CRISPR-Cas gene editing could one day reduce the severity of a serious, food-borne infections that are generally untreatable with antibiotics.
The preclinical findings offer a potential new weapon in the armamentarium against Shiga toxin-producing Escherichia coli (STEC), a bacterium that causes nearly three million acute illnesses worldwide, mostly in children each year.
The precision gene therapy, called EB003, was able to significantly reduce the disease severity in an animal model of STEC infection.
“Although challenges such as dosing, resistance emergence, and delivery optimization remain, our findings provide a compelling foundation for the further development of EB003 and other CRISPR-based therapeutics,” reported Matthieu Galtier, PhD, of the Pasteur Institute in Paris, and co-workers in in Science Translational Medicine.
STEC is a notable public health threat, potentially causing bloody diarrhea that can progress to hemolytic uremic syndrome (HUS), a life-threatening condition with long-term implications.
Many antibiotics can stimulate the production of Shiga toxin, making infections worse and increasing the risk of HUS. They are therefore not generally recommended for infections and patients are typically only treated symptomatically.
Noting the need for strategies to eradicate STEC strains while preventing toxin release, the researchers examined the value of deploying phages—a type of virus that selectively infects and kills bacteria.
Phage-based products are already used to decontaminate food products but, as natural phages do not directly inhibit toxin production, the team investigated whether genetic engineering offered promising solution.
Galtier and colleagues engineered a delivery vector that recognized most clinical isolates from O157:H7 E. coli strains that are commonly associated with human disease, while ensuring a direct shutdown of toxin production.
Previously, the investigators had successfully used a λ phage-derived particle with engineered receptor binding proteins (RBPs) to deliver a base editor to E. coli colonizing the mouse gut.
Building on this λ-based platform, they developed EB003, a rationally engineered antimicrobial that delivers a nonreplicative CRISPR-Cas12 system programmed to target more than 99% of gene variants for stx toxins in E. coli O157 clinical isolates.
The team demonstrated that their nonreplicative CRISPR-Cas antimicrobial was efficient targeting of most clinically relevant O157 E. coli strains in vitro.
It led to a reduction of bacterial load in a mouse gut colonization model, as well as the mitigation of symptoms in an infant rabbit disease model, including reduced toxin production.
![Oral administration of the engineered CRISPR-Cas delivery particle, EB003, significantly reduces disease severity in an infant rabbit model of Shiga toxin-producing E. coli (STEC) infection. [Eligo Bioscience]](https://www.insideprecisionmedicine.com/wp-content/uploads/2026/05/Fig6_Revised-252x300.jpg)
“A key advantage of EB003 over conventional antibiotics is its ability to eliminate O157 STEC without inducing Stx release, a major limitation of current treatments such as fluoroquinolones,” they reported.
“This could translate into a safer therapeutic option, reducing the risk of HUS in patients with a STEC infection.”
The post Precision, Gene-Edited Antimicrobial Targets Difficult-to-Treat <i>E. Coli</i> appeared first on Inside Precision Medicine.
AI Foundation Model Maps Tumor Mutations to Predict Cancer Therapy Response
As genomic sequencing becomes routine in oncology, clinicians are increasingly inundated with tumor mutation data—but often without clear guidance on how to use it. A new artificial intelligence foundation model called MutationProjector may help bridge that gap by transforming complex tumor mutation profiles into clinically meaningful representations that can predict treatment response, classify cancer subtypes, and potentially guide therapeutic decisions across tumor types.
The study, led by researchers at the University of California, San Diego (UCSD) and published in Cancer Discovery, describes how the model was trained on more than 30,000 tumor genomes spanning 10 solid tumor types and then applied to a range of downstream clinical tasks, including prediction of immunotherapy and chemotherapy response.
“Right now, increasingly physicians will order gene sequencing panels, and in a couple of years it’ll be the whole genome,” senior author UCSD’s Trey Ideker, PhD. “What’s interesting is that as prolific as that information is becoming in the cancer clinic, relatively few treatment decisions are actually made based on it.”
According to the paper, FDA-approved targeted therapies currently match only about 8% of patients based on sequencing findings. Yet the average tumor contains roughly 11 distinct genomic alterations identified through clinical sequencing. “The mutations are there,” Ideker said. “People just don’t know what to do with them, or whether they’re informative.”
MutationProjector was designed to address that challenge with architecture inspired by large-scale AI systems used in natural language processing and computer vision. Rather than focusing on single-gene biomarkers, the model attempts to capture broader mutational patterns associated with hallmark cancer pathways.
“We’ve known for years that cancer is really a disease of pathways,” Ideker said. “If you could move beyond individual gene biomarkers and instead understand the overall state of the tumor from its mutations, then you could potentially treat that tumor better.”
The model translates tumor genomic information into a compact coordinate-based representation—similar to a UMAP embedding—that places tumors with similar molecular features near one another. Ideker described the output as an “XY coordinate system for a tumor” that integrates information from thousands of mutations simultaneously.
The approach relies heavily on pretraining, a hallmark of modern foundation models. MutationProjector first learned generalizable representations from large collections of tumor genomic data, even when detailed clinical outcome information was unavailable. Researchers then fine-tuned the model using smaller, highly annotated datasets tied to specific therapeutic responses.
Ideker compared the strategy to language learning. “If I want to learn Italian but I only have one book in Italian, I’m going to learn a lot first from English and Spanish, where I have much more data,” he said. “Then I can transfer that knowledge to Italian much more effectively.”
Similarly, the model learned from tens of thousands of tumor genomes before being tuned to predict outcomes such as immunotherapy response using comparatively small patient cohorts.
The researchers reported that the pretrained embeddings achieved strong performance across multiple downstream tasks, including immunotherapy response prediction, chemotherapy response prediction, and metastasis classification. In one analysis, the model predicted immunotherapy response after fine-tuning on only 94 samples, yet still outperformed competing approaches across independent cohorts.
Importantly, the same pretrained model was applied across multiple tasks rather than building separate models for each application. “These findings reflect the current movement toward general-purpose AI systems capable of addressing a broad range of challenges,” the authors wrote.
The study also highlighted the model’s interpretability. By examining how tumors clustered within the embedding space, researchers identified biologically meaningful patterns associated with treatment response and tumor subtype.
For example, MutationProjector distinguished HPV-positive from HPV-negative tumors in cervical and head and neck cancers, partly through pathway signatures involving apoptosis regulation and Wnt signaling. The model also differentiated basal and luminal subtypes in bladder and breast cancers.
Perhaps more significantly for clinical oncology, the system identified both individual and combinatorial biomarkers linked to immunotherapy outcomes.
Among the findings were associations involving KMT2A and SMARCA4 alterations, which may influence immune response through chromatin remodeling and DNA methylation pathways. The model also captured co-alteration patterns such as KRAS-STK11 and STK11-KEAP1, combinations previously associated with resistance to immunotherapy.
“Those mutations individually may not tell you very much,” Ideker said. “But together, they can become highly predictive of treatment response.”
The authors argue that this ability to model combinatorial mutation patterns could help address one of precision oncology’s central limitations: the difficulty of interpreting rare or co-occurring mutations that do not fit established single-gene biomarker frameworks.
Although the current study focused on 10 solid tumor types, the researchers see considerable room for expansion. Ideker said the team has already scaled training datasets from 30,000 to roughly 300,000 tumor genomes while also increasing the size of the neural network itself.
“We’re now scaling this idea substantially,” he said. “The original work was really a demonstration that the concept works.”
Future directions include expanding into additional tumor types such as pancreatic cancer, prostate cancer, and sarcoma, integrating multimodal data sources including transcriptomics and radiology, and potentially applying the framework to liquid biopsy analysis.
Ideker also envisions the technology eventually supporting molecular tumor boards, where oncologists currently debate difficult treatment decisions based on limited genomic evidence.
“We want to get models like this into those discussions to assist in decision-making,” he said.
The long-term goal, he added, is not simply prediction, but deeper biological understanding. “Once you can interpret the mutational patterns that place tumors into different regions of the map, you start learning the pathways of drug resistance,” Ideker said. “And understanding those pathways is ultimately what drives the next generation of cancer therapeutics.”
The post AI Foundation Model Maps Tumor Mutations to Predict Cancer Therapy Response appeared first on Inside Precision Medicine.