Integrative, and Scalable mental health phenotyping using a knowledge-graph-derived dual-metric framework

This study introduces a scalable mental health phenotyping framework that utilizes a knowledge graph-derived dual-metric system (Cognitive Attention Score and C:ERR), grounded in yogic psychology and lifestyle factors, to provide a multidimensional, validated alternative to traditional symptom-focused diagnostic instruments.

The Gist

Imagine your mental health isn't just a single number on a thermometer, but a complex, living city with traffic patterns, weather systems, and power grids. For decades, doctors have tried to diagnose anxiety and depression by asking you to fill out a survey: "On a scale of 1 to 10, how sad are you today?"

While helpful, this is like trying to understand a city's traffic jam by only asking one driver, "Are you stuck?" It misses the bigger picture: the construction, the rain, the broken traffic lights, and the time of day.

This paper introduces a new way to look at mental health using a "Digital Twin" of your mind, built with two main tools: CAS and C:ERR.

1. The Two New Tools: The "Focus Meter" and the "Emotional Balance Scale"

Instead of just asking "Are you sad?", the researchers built a system that looks at your daily life—how you sleep, what you eat, and how you move—to calculate two scores:

• CAS (Cognitive Attention Score): Think of this as your "Focus Battery." It measures how well your brain is charging up based on your lifestyle. Did you sleep well? Did you eat nutritious food? If yes, your battery is high, and you can focus. If you're running on empty, the battery is low.
• C:ERR (Cognition-to-Emotional-Response Ratio): This is a bit more complex. Imagine your mind is a car. CAS is the engine power. C:ERR is how well the car handles a bumpy road. It measures the ratio between your ability to think clearly and how much your emotions are swerving out of control. A high C:ERR means your emotions are crashing the car; a low C:ERR means you are driving smoothly even when the road is rough.

2. The "Brain Map" (The Knowledge Graph)

The researchers didn't just guess these scores. They built a massive Knowledge Graph (let's call it the "Ceekr-KG").

Imagine a giant, 3D spiderweb where every knot is a piece of information.

• One knot is "Sleep."
• Another is "Anxiety."
• Another is "Back Pain."
• Another is "Low Energy."

In this web, they drew millions of lines connecting these knots based on ancient wisdom from Yogic psychology (which has studied the mind-body connection for thousands of years) and modern science.

• The Magic: When you input your daily habits into this web, the computer doesn't just give you a score; it traces the lines to see how your lack of sleep might be connected to your back pain and your anxiety. It sees the whole city, not just one street.

3. Proving It Works: The "Map vs. Random Dots" Test

To make sure this "Brain Map" wasn't just a lucky guess, they ran a test.

• The Real Map: They used their carefully built web of connections.
• The Fake Map: They took the same dots but scrambled the lines randomly, like taking a map of London and connecting the dots to make a map of a random forest.

The Result: The computer could predict mental health symptoms with 97% accuracy using the Real Map, but it failed miserably with the Fake Map. This proved that the connections they made (the logic of how sleep affects anxiety) were real and meaningful, not just random noise.

4. The "Super-Tool" Effect

They then took their new "Brain Map" and merged it with the world's largest existing medical database (the Clinical Knowledge Graph).

Think of it like upgrading a GPS. The old GPS (standard medical tests) knew the roads. But when they added the new "Ceekr Map" (which knows about lifestyle, diet, and ancient wisdom), the GPS became smarter. It could predict mental health issues better than the old GPS alone, proving that this new way of thinking adds valuable information that doctors were missing.

5. The "Gym for the Mind" Results

Finally, they tested if this system could track improvement. They followed 249 people who were struggling (their "Focus Battery" was low) and put them into three different wellness programs (breathing exercises, coaching, etc.).

• Before: Most people were in the "Low Battery" zone.
• After: After just a few weeks, the average "Focus Battery" jumped up by 11 points.
• The Migration: People didn't just get slightly better; they moved from the "Very Poor" zone to the "Very Good" zone. It was like watching a group of tired hikers suddenly start running up the mountain.

The Big Takeaway

This paper says: Stop just asking people how they feel. Instead, look at how they live.

By combining ancient wisdom about how the mind and body are connected with modern computer science, the researchers created a digital health mirror. This mirror doesn't just tell you "you are sad"; it tells you why (maybe your sleep is off, or your diet is poor) and tracks your progress as you fix those root causes.

It's a shift from treating the symptom (the headache) to understanding the whole system (the stress, the lack of water, and the poor posture) that caused it.

Technical Summary

1. Problem Statement

Current mental health diagnostics for anxiety and depression rely heavily on clinician-administered, symptom-focused questionnaires (e.g., HAM-A, HAM-D, GAD-7). These instruments suffer from several critical limitations:

• Scalability: They are time-intensive (20–25 minutes per administration) and require expert interpretation, hindering large-scale screening and frequent longitudinal monitoring.
• Fragmentation: They often assess anxiety and depression in isolation, ignoring their high comorbidity and the integrated nature of mental health.
• Contextual Blindness: They focus on cognitive and emotional states while neglecting the multidimensional architecture of daily living, including circadian rhythms, nutrition, and lifestyle factors that regulate homeostasis.
• Lack of Integration: Ancient contemplative frameworks (e.g., Yogic psychology) that offer holistic views of mind-body interaction remain disconnected from evidence-based, data-driven medicine due to a lack of rigorous operationalization.

2. Methodology

The authors propose a novel framework that translates Yogic psychological principles into a computational, scalable digital phenotype using Knowledge Graphs (KGs).

A. Dual-Metric Framework (CAS & C:ERR)

Two complementary metrics were derived from Yogic psychology and operationalized within a structured knowledge graph (Ceekr-KG):

1. Cognitive Attention Score (CAS): Quantifies cognitive-attentional state based on self-reported sleep, nutrition, and physical activity.
2. Cognition-to-Emotional-Response Ratio (C:ERR): A weighted composite representing cumulative cognitive and emotional dysregulation.

B. Knowledge Graph Construction (Ceekr-KG)

• Data Source: A knowledge matrix was constructed by translating qualitative Yogic texts into a quantifiable format linking 80 clinical symptoms (physical, cognitive, emotional) to the two metrics.
• Graph Structure: The matrix was converted into a graph with 460 nodes and 151,288 triples.
  • Nodes: 354 discrete CAS levels, 26 continuous C:ERR values, and 80 symptom entities.
  • Edges: Six edge types encoding incremental/decremental relations and polarity.
• Embedding Models: The authors trained and benchmarked five Knowledge Graph Embedding (KGE) models (DisMult, RotatE, ComplEx, RESCAL, SimplE) using the Deep Graph Library (DGL).
  • Optimization: The SimplE model was selected as optimal. Hyperparameter tuning (increasing embedding dimension to 128, adjusting learning rate to 0.5, etc.) significantly improved performance.
  • Validation: A relation-preserving randomization test (shuffling head/tail entities) confirmed that predictive power stems from semantic structure rather than mere graph topology.

C. Integration with Clinical Knowledge Graph (CKG)

To validate clinical relevance, Ceekr-KG was integrated into the independent, large-scale Clinical Knowledge Graph (CKG) (containing ~20M nodes). This created an augmented graph (CKG-CEEKR-KG) to benchmark the new metrics against gold-standard scales (HAM-A, GAD-7, HAM-D).

D. Longitudinal Analysis

The framework was tested on a cohort of 249 individuals (filtered from ~1,840) across three therapeutic programs (AMS, 2M, 3M) with baseline CAS < 64. Improvement was measured via paired t-tests and transition matrices across seven functional bands.

3. Key Results

A. Metric Properties & Correlations

• Distribution: CAS showed a multimodal distribution (peaks at ~30, 60, 85), while C:ERR was right-skewed (mode ~0.9). Both were non-normal.
• Correlation: Strong positive association between CAS and C:ERR (Spearman's $\rho$ = 0.787, $p < 0.0001$).
• Symptom Association: Both metrics showed negative correlations with symptom severity. C:ERR consistently outperformed CAS in predictive strength (e.g., for "low energy," C:ERR $\rho$ = -0.85 vs. CAS $\rho$ = -0.70).
• Complementarity: Ordinal regression showed that a combined model (CAS + C:ERR) achieved the lowest Akaike Information Criterion (AIC) for 71 out of 80 symptoms, proving the metrics provide non-redundant information.

B. Knowledge Graph Performance

• Structural Fidelity: The hyperparameter-tuned SimplE model achieved Hits@1 = 96.9% and MRR = 0.98 on the Ceekr-KG.
• Semantic Validity: Performance on the original graph was 6.9-fold higher than on relation-preserving shuffled controls (Hits@1 ~14%), confirming the model captures genuine semantic constraints.
• Cross-Validation: 5-fold cross-validation yielded consistent results (Hits@1: 95–97%).

C. Diagnostic Benchmarking

• Integrating Ceekr-KG into the CKG systematically improved the predictive scores for anxiety and depression outcomes (HAM-A, GAD-7, HAM-D) compared to the baseline CKG alone.
• This indicates that the Yogic-derived framework encodes clinically relevant semantic information absent in standard biomedical databases.

D. Longitudinal Efficacy

• Improvement: Across all three programs, participants showed significant mean CAS increases (e.g., +11.45 points in the AMS cohort, $p < 0.001$).
• Functional Migration: There was a substantial shift from "Very Poor" and "Poor" functional bands to "Very Good" and "Best" bands.
• 3M Program: The 12-week program achieved complete elimination of severe-risk categories ("Poor" and "Medium" bands were empty at the final measurement) with no observed regression.
• User Validation: 2,029 participants reported high perceived validity (90% agreement), with a distribution significantly skewed toward high engagement ($\chi^2$ test, $p < 0.001$).

4. Key Contributions

1. Operationalization of Ancient Wisdom: Successfully translated complex Yogic psychological concepts into a machine-interpretable, quantitative digital phenotype (CAS/C:ERR).
2. Dual-Metric Framework: Demonstrated that combining a cognitive attention score with an emotional response ratio provides superior, non-redundant predictive power for diverse clinical symptoms compared to single metrics.
3. Knowledge Graph Validation: Proved that the predictive power of the framework arises from its semantic structure (via rigorous randomization controls) rather than topological artifacts.
4. Scalable Phenotyping: Established a method for continuous, scalable mental health monitoring that integrates lifestyle factors, overcoming the limitations of episodic, symptom-only questionnaires.
5. Clinical Integration: Showed that augmenting established biomedical knowledge graphs with these novel metrics improves the inference capabilities of standard diagnostic scales.

5. Significance

This work represents a paradigm shift in mental health assessment by bridging the gap between phenomenological traditions (Yogic psychology) and modern computational medicine.

• Clinical Utility: It offers a scalable, low-cost alternative to expensive neuroimaging or time-consuming clinician interviews, suitable for large-scale screening and real-time longitudinal monitoring.
• Holistic Approach: By capturing the interplay between lifestyle (sleep, diet) and mental states, it addresses the "multidimensional architecture" of mental health often missed by current tools.
• Future Direction: The study validates the use of Knowledge Graphs as a substrate for co-registering historical conceptual frameworks with modern clinical data, paving the way for precision mental health interventions that are both culturally grounded and data-driven.

Limitations noted by authors: Reliance on self-reported data (potential bias), lack of randomized controlled trial (RCT) design for the interventions, and the need for future integration with objective physiological signals (wearables).