Machine learning identifies biological signals linked to emotional hunger

As obesity drug pipelines continue to expand, researchers are trying to better understand why patients respond differently to the same therapies. Although GLP-1 drugs have transformed obesity treatment, variability in response and high discontinuation rates remain significant challenges.

One area attracting increasing interest is emotional hunger, an obesity phenotype associated with emotional and reward-driven eating behaviours. Despite its clinical relevance, the phenotype has remained difficult to study at scale because of the lack of measurable biological markers.

New research from Phenomix Sciences explored whether machine learning and genetic risk scoring could identify biological signals associated with emotional hunger. Presented at the 2026 Pacific Symposium on Biocomputing, the findings demonstrated the feasibility of combining genetic and behavioural data to assess risk in patients with obesity.

Dr Timothy O’Connor, Chief Technology Officer at Phenomix Sciences, spoke with us about the findings and their potential implications for obesity drug development and precision medicine. With more than 20 years of experience spanning bioinformatics, data science and software engineering, O’Connor has previously held roles at Microsoft, Illumina and CareDx. At Phenomix Sciences, he leads development of the company’s machine learning platform, working across clinical and data science teams to analyse biological data associated with obesity phenotypes.

The challenge of defining emotional hunger

Emotional hunger is characterised by eating in response to emotional triggers rather than physiological hunger signals. Patients with this phenotype are often influenced by stress, mood and reward-seeking behaviours. Previous studies have suggested they may respond differently to obesity therapies than other patient groups.

Yet despite its clinical relevance, emotional hunger has remained difficult to quantify biologically.

“The emotional hunger phenotype is harder to define since it sits at the intersection of biology, behaviour and environment,” said O’Connor.

The emotional hunger phenotype is harder to define since it sits at the intersection of biology, behaviour and environment.

Traditional approaches have largely relied on questionnaires and self-reporting tools. While useful, these methods only capture a snapshot in time and do not necessarily reveal underlying biological susceptibility. As a result, emotional hunger has remained difficult to study at scale or incorporate into drug development and clinical research.

The issue is becoming more relevant as obesity treatment options expand and researchers look to better match therapies to specific patient groups.

Machine learning and hidden biological signals

Phenomix Sciences is using machine learning to identify subtle biological patterns associated with emotional hunger. Rather than focusing on a single biomarker, the company’s machine learning genetic risk score (ML-GRS) framework aggregates signals across multiple genetic pathways.

“Machine learning is very valuable since it helps us find subtle signals across multiple genetic variants that would be hard to interpret by themselves,” O’Connor explained. “We can aggregate signals across pathways that are specifically related to mood regulation and reward processing, which allows us to identify patterns associated with emotional hunger,” he added.

Machine learning is very valuable since it helps us find subtle signals across multiple genetic variants that would be hard to interpret by themselves.

The company’s ML-GRS approach combines genetic and behavioural information to build a more complete picture of obesity phenotypes. Genetic variants linked to anxiety, depression and reward-driven eating are analysed alongside targeted behavioural data.

“The ML-GRS model integrates various layers of data,” O’Connor said. “On the genetic side, we look at different variants across pathways linked to anxiety, depression and reward-driven eating to create a score that reflects an individual’s biological susceptibility.”

Importantly, the behavioural component is streamlined rather than relying on lengthy questionnaires.

“The next step is amplifying that signal with targeted behavioural data by focusing on a small number of specific questions rather than a full-length questionnaire,” he explained.

According to O’Connor, this hybrid model allows researchers to distinguish between latent biological risk and active behavioural drivers contributing to weight gain.

Moving beyond single biomarkers

The implications of this work extend well beyond emotional hunger alone. Obesity is widely recognised as a highly heterogeneous disease, with multiple biological pathways contributing to disease progression and treatment response.

For drug developers, this complexity has created major challenges in target identification and patient stratification.

“This research opens new doors for identifying more complex targets,” O’Connor said. “By understanding which genetic pathways are active in specific phenotypes, we can begin to point out what mechanisms are more responsive to treatments.”

Improving clinical trial stratification

One of the most immediate applications could be in early-stage research and clinical development. Obesity trials frequently enrol broad patient populations despite significant biological diversity between individuals. As a result, treatment responses can vary substantially.

By understanding which genetic pathways are active in specific phenotypes, we can begin to point out what mechanisms are more responsive to treatments.

O’Connor believes biological phenotyping could help address this challenge much earlier in the development process.

“This research enables earlier and more precise patient stratification by using biological signals rather than relying solely on observable characteristics or self-reported symptoms,” he said.

“By identifying patients who have a higher chance of having emotional hunger upfront, researchers can define more targeted patient subgroups for clinical trials from the outset.”

Treatment response and patient matching

GLP-1 receptor agonists have changed obesity treatment, but variability in response and high discontinuation rates remain ongoing challenges.

Research suggests emotional hunger patients may respond differently to existing therapies, with some potentially benefiting more from behavioural interventions or treatments such as naltrexone/bupropion than from GLP-1 therapies alone.

O’Connor said the research also reflects a growing focus on precision obesity medicine.

“It’s incredible to see the transformational treatments in the obesity space right now, with many more to come,” he said. “The real challenge is matching the right treatment with the right patient.”

According to O’Connor, this is the central objective of the company’s MyPhenome test and associated research efforts.

It’s incredible to see the transformational treatments in the obesity space right now, with many more to come.

“That’s what we’re achieving at Phenomix. Our MyPhenome test and continued research, like this emotional hunger study, are helping identify distinct biological subtypes of obesity.”

By identifying the biological drivers behind a patient’s condition, clinicians can make more informed treatment decisions and reduce reliance on trial-and-error prescribing.

The implications could also extend into future drug development strategies.

“Looking ahead, these insights have broader implications for drug discovery,” O’Connor explained. “By better defining patient subpopulations, researchers can design and develop therapies that are more targeted and effective from the outset.”

Ultimately, he believes these advances could help deliver more personalised obesity care.

What comes next?

Future research will focus on validating the model in clinical settings.

“The next step would be to validate and refine the emotional hunger predictor in real-world settings,” O’Connor said.

That work will involve increasing clinical datasets and developing collaborations with obesity drug makers, payers and health systems.

As researchers continue exploring more targeted approaches to obesity treatment, identifying biologically meaningful phenotypes such as emotional hunger could become increasingly important for both clinical care and drug development.

Although the research remains at an early stage, the findings demonstrate how machine learning and integrated biomarker strategies could help researchers better understand variation in obesity biology and treatment response.